[
 {
  "id": "model-admet-ai",
  "kind": "model",
  "short_slug": "admet-ai",
  "names": {
   "display": "ADMET-AI",
   "full": null,
   "aliases": [
    "ADMET-AI",
    "admet_ai"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single platform; v1 and v2 are version axes inside one record",
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "ADMET prediction platform returning 41 endpoints per molecule against approved-drug distributions",
  "org": {
   "names": [
    "Stanford University (computer science)",
    "Greenstone Biosciences"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (multitask ADMET ensemble)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code on GitHub with data and models archived in a public repository plus a free web service; licence text not retrieved this round"
  },
  "first_release": {
   "date": "2023-12-28",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint posted (registry record)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1101/2023.12.28.573531",
   "verification": "verified",
   "note": "The site recorded 2024, the journal year"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-12-28",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2023.12.28.573531",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2023-12-28",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2023.12.28.573531",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2024-06-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal publication (registry record)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1093/bioinformatics/btae416",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2026-02-17",
    "precision": "day",
    "year_hint": null,
    "basis": "v2.0.0 release (all models retrained)",
    "event_kind": "new-version",
    "timestamp_utc": null,
    "source_url": "https://github.com/swansonk14/admet_ai/releases",
    "verification": "verified",
    "note": "No release-notes text was retrieved for v2.0.0; the description comes from the README"
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-02-22",
   "precision": "day",
   "year_hint": null,
   "basis": "v2.0.1 patch (latest tag)",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://github.com/swansonk14/admet_ai/releases",
   "verification": "verified",
   "description": "v2.0.0 (2026-02-17) retrained all models from scratch so predictions differ from the v1 web service; v2.0.1 fixed MPS/CPU handling and dropped a dataset-library dependency. Per the package README the web service still runs v1",
   "note": null
  },
  "status": "active",
  "status_note": "Major version and patch released within 12 months",
  "what_it_does": "Returns predictions for 41 absorption, distribution, metabolism, excretion and toxicity endpoints per molecule and compares them with the distribution of approved drugs, aimed at large-scale library screening (the paper reports a few hours for a million molecules locally). The model itself is not new: it is a descriptor-augmented multitask ensemble of the previous Chemprop generation, and its value is engineering and coverage; unlike pretrained foundation models such as MolE and MiniMol it does no large-scale pretraining.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "41 ADMET datasets (10 regression, 31 classification) from a public therapeutics dataset library, 22 of which form its official ADMET leaderboard; five models per dataset as an ensemble",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11226862/"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11226862/"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11226862/",
    "confidence": "high",
    "second_hand": false,
    "note": "Submitted to that library's ADMET group board; the paper reports the best average rank, but a full sweep on 2026-11-16 found only one submission labelled with this model, covering a single task and placed last (value about 25x the best, a suspected unit error), so \"best average rank across 22 datasets\" has no counterpart on the current board"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own throughput measurement",
    "category": "own-task-set",
    "detail": "The paper reports the time to predict a million molecules locally; an engineering figure, not an accuracy evaluation",
    "source_type": "self"
   },
   {
    "label": "Comparison with approved-drug distributions",
    "category": "own-task-set",
    "detail": "Places predictions on the distribution of approved drugs as percentiles; an interpretive output",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a free web service plus open local inference covering 41 ADMET endpoints at once, with the best average rank on the public ADMET group board as it stood at publication.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation of the platform was found, and the current board holds no entry supporting the best-average-rank claim",
  "background": {
   "org_profile": "The authors come from a US university computer science department and also carry the affiliation of a biotechnology company in the university's research park; financial-data archives record the company as founded in 2021, working on induced pluripotent stem cells and organoids with machine learning for discovery and safety prediction, and claiming a large stem-cell biobank.",
   "funding": "Public archives list three venture investors in the company; academic funding is not itemised in the paper.",
   "products_partnerships": "The product is a free web service plus open local inference; on 2026-06-19 the company announced a collaboration with a semiconductor company to process its stem-cell biobank data on that company's compute (press release wording).",
   "related_models": [
    "model-chemprop-v2",
    "model-minimol",
    "model-mole"
   ]
  },
  "reading_notes": [
   "\"Best average rank\" aggregates 22 datasets as of publication and does not mean first on each endpoint; no matching entry exists on the current board, so any citation needs a date and source qualifier.",
   "The only submission on the current board labelled with this model covers one task, sits last and carries a suspected unit error; it cannot be read as its present standing.",
   "It shows a well-tuned graph-network ensemble with descriptors remains strong on ADMET, so pretrained foundation models do not hold a stable edge.",
   "From v2.0.0 all models were retrained while the web service still runs v1, so the two give different predictions and comparisons must state the version."
  ],
  "caveats": [
   "The paper's board standing does not match the board as of 2026-11-16; the published statement is corrected.",
   "The public ADMET board uses public test sets, so overfitting and differing submission protocols matter.",
   "No wet lab and no private data; all training data is public.",
   "No release-notes text was retrieved for v2.0.0; the version description comes from the package documentation."
  ],
  "sources": [
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11226862/",
    "title": "Journal full text (public archive)",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://doi.org/10.1093/bioinformatics/btae416",
    "title": "Journal paper 2024",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/swansonk14/admet_ai/releases",
    "title": "Release list",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://pypi.org/project/admet-ai",
    "title": "Package registry page",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "title": "Public ADMET group board overview (checked 2026-11-16)",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "Best average rank across 22 datasets",
    "now": "The claim holds only as of the paper; the only submission labelled with this model on the current board is on a single task and in last place (suspected unit error); no matching entry",
    "source_url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2024",
    "now": "2023-12-28 (preprint posted)",
    "source_url": "https://api.crossref.org/works/10.1101/2023.12.28.573531",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2026-02-17（v2.0.0）",
    "now": "2026-02-22 (v2.0.1, latest tag), with the full v2 retraining also recorded",
    "source_url": "https://github.com/swansonk14/admet_ai/releases",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.code_release",
   "open_status.code_license",
   "open_status.weights_license",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-admet-ai/",
   "en": "/bio/benchmark/models/admet-ai/"
  },
  "names_full_null_reason": "The name is itself an acronym compound; the paper gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-af2-multimer",
  "kind": "model",
  "short_slug": "af2-multimer",
  "names": {
   "display": "AF2-Multimer",
   "full": "AlphaFold2-Multimer",
   "full_null_reason": null,
   "aliases": [
    "AlphaFold2-Multimer",
    "AlphaFold-Multimer",
    "AF2-Multimer",
    "AF-Multimer"
   ],
   "ambiguous_aliases": [
    "AlphaFold2",
    "AlphaFold"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model record; the multimer v3 weights of v2.3 are a milestone, not a family member",
  "domain": "structure",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Multi-chain fine-tune of AF2; the classic protein-complex baseline",
  "org": {
   "names": [
    "Google DeepMind"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Co-folding (protein complexes only)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "CC BY 4.0 (per site record; official source governs)",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code and parameters ship with the AlphaFold repository; licence terms per the official page, text not re-verified this round"
  },
  "first_release": {
   "date": "2021-10-04",
   "precision": "day",
   "event_kind": "paper-revision",
   "basis": "bioRxiv preprint 'Protein complex prediction with AlphaFold-Multimer' v1",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2021.10.04.463034",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2021-10-04",
    "precision": "day",
    "event_kind": "paper-revision",
    "basis": "bioRxiv preprint v1, earliest provable public evidence",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2021.10.04.463034",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2021-10-04",
    "precision": "day",
    "event_kind": "paper-revision",
    "basis": "bioRxiv v1; no journal version of the same paper found",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2021.10.04.463034",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2022-12-13",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "AlphaFold v2.3.0 shipped the multimer v3 weights",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphafold/releases",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2022-12-13",
   "precision": "day",
   "event_kind": "new-weights",
   "basis": "AlphaFold v2.3.0 released the multimer v3 weights",
   "timestamp_utc": null,
   "source_url": "https://github.com/google-deepmind/alphafold/releases",
   "verification": "verified",
   "note": null
  },
  "status": "superseded",
  "status_note": "Complex prediction superseded by AlphaFold 3; kept as a baseline",
  "what_it_does": "Fine-tunes AF2 on multi-chain data to output protein-complex structures with an interface confidence (ipTM). Proteins only, no ligands or nucleic acids, which is what AlphaFold 3 later superseded; large-scale sampling methods such as MassiveFold use it as their engine.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB multimers, UniRef and similar public databases; described on the site as general knowledge, not checked item by item",
    "source_url": "https://doi.org/10.1101/2021.10.04.463034"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The preprint lists public databases only and discloses no licensed data",
    "source_url": "https://doi.org/10.1101/2021.10.04.463034"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-22-capri",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "CASP15-CAPRI official assessment, assessor-run yardstick baseline",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/37905971/",
    "confidence": "medium",
    "verification": "partial",
    "note": "An assessor-run yardstick baseline inside the official assessment, with no separate figure; this is why the model leaves the zero-pairing list this round"
   }
  ],
  "other_evaluations": [
   {
    "label": "Seventeen-heterodimer no-template benchmark",
    "category": "public-benchmark-not-on-site",
    "detail": "Interface accuracy on seventeen heterodimers in a no-template setting; not among the site's 44 benchmarks (developer-reported)",
    "source_type": "self"
   },
   {
    "label": "Large recent-PDB complex set",
    "category": "own-task-set",
    "detail": "Developer-built set of 4,446 recent complexes with non-redundant low-template interfaces, compared against AF2 run with a linker (self-reported)",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: interface-prediction success is substantially higher than AlphaFold run with a linker, on both heteromeric and homomeric interfaces, including a no-template setting (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "It appears mainly as a third-party baseline: the CASP15-CAPRI assessors used it as their yardstick and the NeuralPLexer3 paper re-ran it as a standard comparison.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Google DeepMind is Alphabet's AI lab in London; DeepMind was founded in 2010, acquired by Google in 2014 and merged with Google Brain into Google DeepMind in 2023-04.",
   "funding": null,
   "funding_null_reason": "Funded from Alphabet's internal budget, no separate rounds",
   "products_partnerships": "Successor to AF2 and predecessor of AlphaFold 3 and IsoDDE; MassiveFold runs large-scale sampling on the AF2 line.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3",
    "model-massivefold",
    "model-isodde"
   ]
  },
  "reading_notes": [
   "Its value is as a baseline: a new model's gain over it on complex interfaces is the minimum bar for progress.",
   "Comparison figures in third-party papers are those authors' own re-runs, not developer self-reports, and should not be pooled.",
   "It has no ligand or nucleic-acid capability, so it cannot anchor small-molecule or RNA comparisons.",
   "After 2022-12 only fixes and cleanup commits followed; none counts as a new model generation."
  ],
  "caveats": [
   "It leaves the zero-pairing list this round as an official CAPRI assessment baseline, though that pairing is only medium confidence.",
   "Figures on its own task set and on off-site benchmarks are developer-reported, with no independent replication found.",
   "Training data is listed as general knowledge; per-database sizes are not verified.",
   "Licence terms (code Apache-2.0, parameters CC BY 4.0) follow the older record; the official page governs."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2021.10.04.463034",
    "title": "AlphaFold-Multimer bioRxiv preprint",
    "kind": "preprint",
    "date": {
     "date": "2021-10-04",
     "precision": "day",
     "event_kind": "paper-revision",
     "basis": "preprint v1",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1101/2021.10.04.463034",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://github.com/google-deepmind/alphafold/releases",
    "title": "AlphaFold repository releases",
    "kind": "repository",
    "date": {
     "date": "2022-12-13",
     "precision": "day",
     "event_kind": "new-weights",
     "basis": "v2.3.0",
     "timestamp_utc": null,
     "source_url": "https://github.com/google-deepmind/alphafold/releases",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://pubmed.ncbi.nlm.nih.gov/37905971/",
    "title": "CASP15-CAPRI official assessment paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2412.10743",
    "title": "NeuralPLexer3 paper (third-party comparison)",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "Zero pairings: no site benchmark pairing",
    "now": "Counted as an official CAPRI assessment baseline; leaves the zero-pairing list this round",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/37905971/",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2021-10 (month only)",
    "now": "2021-10-04 (bioRxiv v1)",
    "source_url": "https://doi.org/10.1101/2021.10.04.463034",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[0]",
   "training_data.a_public",
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-af2-multimer/",
   "en": "/bio/benchmark/models/af2-multimer/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-aido-cell",
  "kind": "model",
  "short_slug": "aido-cell",
  "names": {
   "display": "AIDO Cell",
   "full": null,
   "aliases": [
    "AIDO Cell 1.0",
    "AIDO.Cell",
    "GB.Cell"
   ],
   "ambiguous_aliases": [
    "AIDO"
   ]
  },
  "family": "AIDO",
  "family_null_reason": null,
  "domain": "single-cell",
  "domains_secondary": [
   "genomics-rna",
   "structure"
  ],
  "one_liner": "Multi-scale, stateful virtual-cell world model, currently with two prototype cell types",
  "org": {
   "names": [
    "GenBio AI"
   ],
   "global_entity_ids": [
    "genbio-ai"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Virtual cell / multi-scale world model",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "This version releases no weights and its white paper is on request; the previous generation's single-cell weights are on the model hub under the developer's own community licence"
  },
  "first_release": {
   "date": "2026-08-18",
   "precision": "day",
   "year_hint": null,
   "basis": "Official launch post with same-day press coverage",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://genbio.ai/aido-cell-simulator/",
   "verification": "verified",
   "note": "One outlet gives 08-19 while a same-day report gives 08-18, which is used; the previous generation's single-cell weights appeared in 2024-11 and are a different product"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-08-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Official launch post",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://genbio.ai/aido-cell-simulator/",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-08-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Official launch post",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://genbio.ai/aido-cell-simulator/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-08-18",
   "precision": "day",
   "year_hint": null,
   "basis": "No sourced substantive event since launch",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://genbio.ai/aido-cell-simulator/",
   "verification": "verified",
   "description": "Access is still a waiting list with the early-access programme in preparation and the full white paper available on request; a 2026-07-25 change to the previous generation's repository belongs to a different product line",
   "note": null
  },
  "status": "product-only",
  "status_note": "No public weights or reproducible materials, waiting list only",
  "what_it_does": "Simulates DNA, RNA, protein and whole-cell behaviour in one shared state where perturbations accumulate and can be applied in sequence, currently for two prototype cell types only. Unlike transcriptome-perturbation models such as STATE and Lingshu-Cell, its comparators span structure, regulation and splicing models including AlphaFold 3 and AlphaGenome.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "This version's training data composition is undisclosed; the previous generation used public single-cell resources and is a different product",
    "source_url": "https://genbio.ai/aido-cell-simulator/"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://genbio.ai/aido-cell-simulator/"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://genbio.ai/aido-cell-simulator/"
   },
   "d_wetlab_validation_only": {
    "value": "not-disclosed",
    "detail": "Public material shows only a replay of a known drug mechanism, with no prospective wet-lab work mentioned",
    "source_url": "https://genbio.ai/aido-cell-simulator/"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-09-scib",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "self",
    "evidence_url": "https://huggingface.co/genbio-ai/GB.Cell-100M",
    "confidence": "medium",
    "second_hand": false,
    "note": "The previous generation's model card claims leading zero-shot clustering and related results without naming that evaluation or giving figures; treated as unconfirmed and not counted"
   }
  ],
  "other_evaluations": [
   {
    "label": "Vendor's own virtual-cell evaluation",
    "category": "public-benchmark-not-on-site",
    "detail": "Designed, split and run by the developer across perturbation, knockout, structure, splicing and regulation, with self-chosen comparators and metrics; per-item values appear only in the on-request white paper",
    "source_type": "self"
   },
   {
    "label": "Replay of a known drug mechanism",
    "category": "case-study",
    "detail": "Demonstrates a known kinase inhibitor's action in a leukaemia cell line; a case study, not prospective validation",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report it is the only system covering all five task classes and matching or beating the strongest specialist baselines on most metrics of their own evaluation.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The evaluation is vendor-built and vendor-run with no third-party reproduction, and neither weights nor white paper are public",
  "background": {
   "org_profile": "The developer is an AI-native biology company headquartered in California with satellite labs in Europe and the Middle East, first presenting its digital-organism programme in 2024-12; founding year and headcount are not public.",
   "funding": "Press releases say only that it is backed by leading investors; round sizes and lead investors are not public.",
   "products_partnerships": "A first phase in 2024-12 covered six foundation models across DNA, RNA and single cells, and this virtual-cell product followed; researchers from several universities took part and no pharmaceutical partnerships are public.",
   "related_models": [
    "model-state",
    "model-lingshu-cell",
    "model-scgpt-geneformer"
   ]
  },
  "reading_notes": [
   "A vendor-built, vendor-run evaluation is not a blind test: tasks, splits and comparators are the vendor's and per-item values are not public.",
   "A count aggregated across structure, splicing, regulation and perturbation cannot be converted into usability on any one task class.",
   "Only two prototype cell types exist and no prospective wet-lab validation has been seen.",
   "The previous-generation single-cell model and this version are different products and their figures must not be mixed."
  ],
  "caveats": [
   "Under this round's relation rules it enters the zero-pairing list: the model card's claim names no benchmark and gives no figures.",
   "Weights and full materials are not public, so no claim can be checked externally.",
   "Two dates circulate for the launch; this record takes the one from same-day coverage."
  ],
  "sources": [
   {
    "url": "https://genbio.ai/aido-cell-simulator/",
    "title": "Official product page",
    "kind": "official-page",
    "date": null
   },
   {
    "url": "https://www.statnews.com/2026/08/18/david-baker-genbio-ai-new-virtual-model-unveiled-aido-cell/",
    "title": "Press coverage (launch day)",
    "kind": "news",
    "date": "2026-08-18"
   },
   {
    "url": "https://huggingface.co/genbio-ai/GB.Cell-100M",
    "title": "Previous-generation model card",
    "kind": "model-hub",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2024.11.28.625303",
    "title": "Previous-generation preprint",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "Self-reported lead on this single-cell integration benchmark (self)",
    "now": "Changed to unconfirmed: the model card does not name the benchmark, gives no numbers, and refers to the previous-generation product",
    "source_url": "https://huggingface.co/genbio-ai/GB.Cell-100M",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "training_data.a_public",
   "training_data.d_wetlab_validation_only",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-aido-cell/",
   "en": "/bio/benchmark/models/aido-cell/"
  },
  "names_full_null_reason": "The name is a product name; the official page gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 1
   }
  },
  "global_entity_null_reason": "No model page in the site-wide entity library; only the developer organisation has one",
  "status_head": null
 },
 {
  "id": "model-alphafold3",
  "kind": "model",
  "short_slug": "alphafold3",
  "names": {
   "display": "AlphaFold 3",
   "full": null,
   "full_null_reason": "No expanded name; official materials use AlphaFold 3",
   "aliases": [
    "AlphaFold 3",
    "AlphaFold3",
    "AF3"
   ],
   "ambiguous_aliases": [
    "AlphaFold",
    "AlphaFold Server"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model record; the earlier complex model AF2-Multimer has its own record",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Biomolecular co-folding model that later co-folding models use as the reference baseline",
  "org": {
   "names": [
    "Google DeepMind",
    "Isomorphic Labs"
   ],
   "global_entity_ids": [
    "isomorphic-labs"
   ]
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Co-folding (protein / nucleic acid / small molecule / ion)",
  "open_status": {
   "code": "partial",
   "weights": "on-request",
   "weights_license": "on request, non-commercial per earlier note, not re-verified",
   "code_license": "Apache-2.0 from v3.0.3 (2026-06-09)",
   "license_changed": {
    "date": "2026-06-09",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "v3.0.3 release notes relicensed the code to Apache-2.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphafold3/releases",
    "verification": "verified",
    "note": null
   },
   "note": "Code and weights terms must be stated separately: code is Apache-2.0, weights are on request and their terms are still unconfirmed (unverified)"
  },
  "first_release": {
   "date": "2024-05-08",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Journal paper online with the official server the same day; code and weights followed with v3.0.0 on 2024-11-11",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41586-024-07487-w",
   "verification": "partial",
   "note": "No primary metadata for this date was saved this round, so it is marked unverified"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": {
    "date": "2024-05-08",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Journal paper online",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41586-024-07487-w",
    "verification": "partial",
    "note": null
   },
   "code_release": {
    "date": "2024-11-11",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Code public with GitHub v3.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphafold3/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2024-11-11",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights available on request with v3.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphafold3/releases",
    "verification": "partial",
    "note": "Terms not re-verified"
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-05-08",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Official server launch announcement",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41586-024-07487-w",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-30",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release v3.0.5",
   "description": "v3.0.5 speeds up search and MSA featurisation and adds a CPU flag and a shard helper; v3.0.3 (2026-06-09) relicensed the code to Apache-2.0",
   "timestamp_utc": null,
   "source_url": "https://github.com/google-deepmind/alphafold3/releases",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Code releases within 12 months, no new generation",
  "what_it_does": "A diffusion module generates all-atom coordinates directly, covering protein-ligand, protein-nucleic-acid and antibody-antigen complexes in one model. It is the de facto co-folding standard: open efforts such as Boltz-1/1x, Chai-1, Protenix and OpenFold3 set out to reproduce or approach it.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public PDB plus distillation data; the paper gives no per-database sizes",
    "source_url": "https://doi.org/10.1038/s41586-024-07487-w"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The developers report no wet-lab work",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "CASP15 public RNA targets",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11168924/",
    "confidence": "high",
    "verification": "partial",
    "note": "The paper reuses released targets post hoc; in CASP16 it appears only as a component of other groups' pipelines, with no entry of its own"
   },
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "PoseBusters V1 (428), official set",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11168924/",
    "confidence": "high",
    "verification": "verified",
    "note": "V1 is a self-report on the official set; V2 figures are third-party recomputations"
   },
   {
    "benchmark_id": "bench-12-dude",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": null,
    "source_url": null,
    "confidence": "low",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): one second-hand abstract mentions the family, with no values, and the benchmark paper does not evaluate it"
   },
   {
    "benchmark_id": "bench-15-cameo",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "operator-run complex and ligand baseline, 2024-01 to 2025-06",
    "source_url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12750032/",
    "confidence": "medium",
    "verification": "partial",
    "note": "Run by the operators as a baseline, not a developer submission"
   },
   {
    "benchmark_id": "bench-22-capri",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": "CASP16 complexes",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12750027/",
    "confidence": "medium",
    "verification": "unverified",
    "note": "The official assessment has no entry of its own; the claim is second-hand (unverified)"
   },
   {
    "benchmark_id": "bench-25-rnapuzzles",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "CASP15 RNA, quoted in another paper",
    "source_url": "https://arxiv.org/abs/2412.10743",
    "confidence": "medium",
    "verification": "partial",
    "note": "The figure comes from its own paper quoted by a third party, not from that challenge"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "Runs N' Poses, best@5",
    "source_url": "https://arxiv.org/pdf/2510.24670",
    "confidence": "medium",
    "verification": "partial",
    "note": "Quoted second-hand from a third-party paper, not a developer self-report"
   },
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer leaderboard",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "high",
    "verification": "verified",
    "note": null
   },
   {
    "benchmark_id": "bench-33-posex",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "self-docking / cross-docking, benchmark authors' run",
    "source_url": "https://arxiv.org/html/2505.01700v3",
    "confidence": "medium",
    "verification": "partial",
    "note": "Re-run by the benchmark's authors under the same protocol"
   },
   {
    "benchmark_id": "bench-34-abbibench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "registered interface-score entry",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): the entry is an interface-score correlation read off a paper figure, not structure accuracy, and was not re-run"
   }
  ],
  "other_evaluations": [
   {
    "label": "Third-party prospective crystal-structure evaluation",
    "category": "third-party-prospective-evaluation",
    "detail": "A prospective comparison on about 551-557 newly solved complexes: good pose quality, but confidence is not fit for virtual-screening ranking (preprint plus second-hand)",
    "source_type": "third"
   },
   {
    "label": "Role as a control in 2026 boards",
    "category": "comparison-vs-baseline-only",
    "detail": "Most new models' papers list it as the main control, while the developers rarely report new evaluations",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: protein-ligand docking is significantly better than classical docking tools, with protein, nucleic acid, small-molecule and ion complexes unified in one model.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Several third-party papers recomputed it under stricter protocols and a maintainer leaderboard includes it; leakage-controlled splits show accuracy falling as similarity drops, indicating memorisation.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The model was built by a big-tech AI lab together with its sister company, which was spun out of the same parent in 2021, is headquartered in London and positions itself as an AI drug-design company; it reports over 350 staff across London, Cambridge (Massachusetts) and Lausanne (2026-05 reporting).",
   "funding": "2025-03-31 first external round of $600M led by Thrive Capital with GV and Alphabet; 2026-05-12 a $2.1B Series B again led by Thrive Capital, adding MGX, Temasek, CapitalG and a UK sovereign AI fund; valuation undisclosed.",
   "funding_null_reason": null,
   "products_partnerships": "2024-01 deals with two large pharma companies totalling $82.5M upfront and up to about $2.9B in milestones, excluding royalties. The internal engine IsoDDE (disclosed 2026-02-10) extends it; the official blog gives no numbers page.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-af2-multimer",
    "model-openfold3",
    "model-protenix",
    "model-boltz-1-1x",
    "model-chai-1",
    "model-isodde"
   ]
  },
  "reading_notes": [
   "Its docking results differ widely across papers because PoseBusters V1 and V2, sample counts and pose-selection rules differ; the numbers are not comparable across papers.",
   "All co-folding models degrade once results are stratified by similarity to training data, so its high scores should be read per similarity bin.",
   "In third-party prospective evaluations pose quality is good, but confidence is not sufficient for virtual-screening ranking on its own.",
   "The code is Apache-2.0 while the weights remain on request with unconfirmed terms, so commercial usability turns on the weights side."
  ],
  "caveats": [
   "Code and weights terms are separate: the weights terms are unconfirmed and must not be inferred from the code licence (unverified).",
   "The first-release date lacks primary metadata and is marked partially verified.",
   "Two pairings rest only on the site's own pages and must display as unverified.",
   "Challenges whose official assessments contain no entry of its own are not counted as established relations here."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41586-024-07487-w",
    "title": "The model's original journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11168924/",
    "title": "Paper full text (public archive)",
    "kind": "paper",
    "date": null
   },
   {
    "url": "https://github.com/google-deepmind/alphafold3/releases",
    "title": "Official repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2412.10743",
    "title": "One of the third-party recomputation papers",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/pdf/2510.24670",
    "title": "A second third-party recomputation paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12750032/",
    "title": "Baseline paper by the continuous-evaluation operators",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://github.com/BEAM-Labs/FoldBench",
    "title": "Maintainer leaderboard repository",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2505.01700v3",
    "title": "Re-run paper by the benchmark's authors",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://isomorphiclabs.com/articles/isomorphic-labs-announces-600m-external-investment-round",
    "title": "Official announcement of the first external round",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "open_status",
    "was": "Code and weights are both limited to non-commercial academic request",
    "now": "Code is Apache-2.0 from v3.0.3 (2026-06-09); weights are on request with their own, still unconfirmed terms",
    "source_url": "https://github.com/google-deepmind/alphafold3/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "The first release was treated as fully verified from the journal date",
    "now": "The date stands, but with no primary metadata this round it is downgraded to partially verified",
    "source_url": "https://doi.org/10.1038/s41586-024-07487-w",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Third-party recomputations and developer self-reports were merged into one cell",
    "now": "Split across the seven relation types: official board, self-report on the official protocol, third-party re-run and unconfirmed each stand alone",
    "source_url": null,
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "open_status.weights_license",
   "release_milestones.weights_release",
   "benchmarks_reported[2]",
   "benchmarks_reported[4]",
   "benchmarks_reported[9]"
  ],
  "as_of": "2026-11-16",
  "global_entity": "alphafold3",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-alphafold3/",
   "en": "/bio/benchmark/models/alphafold3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-alphagenome",
  "kind": "model",
  "short_slug": "alphagenome",
  "names": {
   "display": "AlphaGenome",
   "full": null,
   "aliases": [
    "AlphaGenome"
   ],
   "ambiguous_aliases": [
    "AlphaGenome 2"
   ]
  },
  "family": null,
  "family_null_reason": "The successor AlphaGenome 2 (2026-11-11) is recorded as the latest event here; no family entry",
  "domain": "genomics-rna",
  "domains_secondary": [],
  "one_liner": "Sequence-to-function model reading 1 Mb of DNA for base-resolution regulatory tracks and variant effects",
  "org": {
   "names": [
    "Google DeepMind"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Genome sequence-to-function model / variant effect prediction",
  "open_status": {
   "code": "open",
   "weights": "non-commercial",
   "weights_license": "non-commercial terms, acceptance required",
   "code_license": "documentation and example code CC-BY-4.0",
   "license_changed": null,
   "note": "Weights and inference code public from 2026-01-28; API open for non-commercial use; the successor's weights are not mentioned"
  },
  "first_release": {
   "date": "2025-06-25",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer blog announcement with API preview (preprint 2025-06-27)",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/",
   "verification": "partial",
   "note": "The announcement date was not directly re-verified and comes from aggregators"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-06-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Blog announcement and API preview",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-06-27",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv v1 (registered date)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2025.06.25.661532",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-01-28",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal online publication",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41586-025-10014-0",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2026-01-28",
    "precision": "day",
    "year_hint": null,
    "basis": "Inference code released with the journal paper",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphagenome_research",
    "verification": "verified",
    "note": "The research repository also has v0.1.0 (2026-02-20) and v0.3.0 (2026-07-14)"
   },
   "weights_release": {
    "date": "2026-01-28",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights released under non-commercial terms",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41586-025-10014-0",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-06-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer blog",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-11-11",
   "precision": "day",
   "year_hint": null,
   "basis": "Successor model AlphaGenome 2 launch and API v2.0.0",
   "event_kind": "successor-model",
   "timestamp_utc": null,
   "source_url": "https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/",
   "verification": "verified",
   "description": "AlphaGenome 2 launched: a conditional model predicting one tissue or cell type at a time, reported by the developers as trained on over a hundred million single-cell profiles; available by API for non-commercial use with a commercial private preview, weights not mentioned",
   "note": "The published record did not include this event"
  },
  "status": "superseded",
  "status_note": "The successor AlphaGenome 2 has been released",
  "what_it_does": "Reads 1 Mb of DNA and predicts expression, transcription initiation, chromatin accessibility, histone marks, transcription-factor binding, contact maps and splice-site usage at base resolution, scoring non-coding variants by the difference between reference and alternate predictions. Compared with Borzoi it lengthens the input and merges modalities into one model; unlike self-supervised DNA language models such as Evo 2 and NTv3 it is supervised on functional-genomics tracks.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public functional-genomics and variant resources; the paper states no new raw experimental data was generated",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12851941/"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12851941/"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The reporter-assay and variant validations in the paper use public data only",
    "source_url": "https://doi.org/10.1038/s41586-025-10014-0"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-08-enformer",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.1038/s41586-025-10014-0",
    "confidence": "high",
    "second_hand": false,
    "note": "The predecessor's track test set; comparisons are against the strongest external model per task, not a common leaderboard"
   },
   {
    "benchmark_id": "bench-11-clinvar",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.1038/s41586-025-10014-0",
    "confidence": "medium",
    "second_hand": false,
    "note": "A developer-built variant-effect suite covering splice-related classes only"
   },
   {
    "benchmark_id": "bench-37-traitgym",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.1038/s41586-025-10014-0",
    "confidence": "medium",
    "second_hand": false,
    "note": "Uses the benchmark's variant sets but reports enrichment fractions rather than the leaderboard metric, so no comparable score exists"
   }
  ],
  "other_evaluations": [
   {
    "label": "QTL and causality tasks",
    "category": "own-task-set",
    "detail": "Direction and coefficient correlation for expression and accessibility QTLs, plus a supervised classifier trained on multimodal scores, all developer-defined",
    "source_type": "self"
   },
   {
    "label": "Public reporter-assay datasets",
    "category": "public-benchmark-not-on-site",
    "detail": "Splicing and regulatory evaluations on two public MPRA datasets, outside the 44",
    "source_type": "self"
   },
   {
    "label": "Third-party probing comparison (2026-04)",
    "category": "comparison-vs-baseline-only",
    "detail": "A third-party study claims probes trained on another genome model's embeddings beat this model; those are probes rather than zero-shot runs and are not directly comparable",
    "source_type": "third"
   },
   {
    "label": "BEND",
    "category": "public-benchmark-not-on-site",
    "detail": "The original paper does not report this benchmark",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one model matches or beats the strongest external model per task on most track and variant-effect evaluations, with multimodal prediction in a single 1 Mb context.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation under a site benchmark protocol was found even after the weights were released; existing third-party work uses a probing setup",
  "background": {
   "org_profile": "A London AI lab inside a US-listed technology group, founded in 2010, acquired in 2014 and merged with another group research unit in 2023-04. Headcount framings differ: a 2025-09 media report cites about 5,600 (a wider remit) while a data firm counts 3,678 for its UK entity (2025).",
   "funding": "Funded from the group's R&D budget; a sister AI drug-design company in the same group announced a $600M first external round on 2025-03-31.",
   "products_partnerships": "A sibling co-folding model AlphaFold 3; two earlier sequence-to-function models, one of which (Borzoi) is both predecessor and comparator; at API launch the developers said thousands of scientists were testing it; the group also has the single-cell model C2S-Scale with a US university.",
   "related_models": [
    "model-borzoi",
    "model-alphafold3",
    "model-c2s-scale",
    "model-evo-2",
    "model-ntv3"
   ]
  },
  "reading_notes": [
   "All win counts are self-reported and measured against the strongest external model per task, not a rank on a common third-party board.",
   "The 2025-06 preview and the journal version count differently; always state which version is cited.",
   "The ClinVar side covers splice-related classes only and does not show general pathogenicity classification is usable.",
   "Long-range regulation (enhancer-gene links beyond 100 kb) is an acknowledged weakness, and only TraitGym's variant sets were used."
  ],
  "caveats": [
   "The preview and journal versions count differently and must not be mixed.",
   "Weights carry non-commercial terms, and the model card and paper conflict on whether one tissue dataset is in the released version.",
   "All site benchmark entries are self-reported; no third-party figures are citable.",
   "The successor is API-only, with weights and evaluation detail undisclosed."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41586-025-10014-0",
    "title": "Journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12851941/",
    "title": "Journal full text (public archive)",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/",
    "title": "Developer blog (preview)",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/",
    "title": "Developer blog (successor model)",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://github.com/google-deepmind/alphagenome_research",
    "title": "Code and weights repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "No 2026 event",
    "now": "2026-11-11 successor model AlphaGenome 2 released (API v2.0.0)",
    "source_url": "https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/",
    "date_found": "2026-11-16"
   },
   {
    "field": "status",
    "was": "Active",
    "now": "Superseded by a successor",
    "source_url": "https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.official_announcement",
   "training_data.a_public"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-alphagenome/",
   "en": "/bio/benchmark/models/alphagenome/"
  },
  "names_full_null_reason": "The developer uses the model name only",
  "n_benchmarks_reported": {
   "counted": 3,
   "by_relation": {
    "self-reported-official-protocol": 1,
    "self-reported-modified-protocol": 2
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-alphamissense",
  "kind": "model",
  "short_slug": "alphamissense",
  "names": {
   "display": "AlphaMissense",
   "full": null,
   "aliases": [
    "AlphaMissense"
   ],
   "ambiguous_aliases": [
    "AlphaFold"
   ]
  },
  "family": null,
  "family_null_reason": "Record covers the missense model only; the sibling genomics and protein-design models have their own records",
  "domain": "genomics-rna",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Structure-augmented missense pathogenicity model with a large public variant catalogue",
  "org": {
   "names": [
    "Google DeepMind"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Variant pathogenicity prediction (structure-augmented protein LM)",
  "open_status": {
   "code": "open",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": {
    "date": "2024-03-13",
    "precision": "day",
    "year_hint": null,
    "basis": "Predictions-database licence update (last repository commit)",
    "event_kind": "documentation-update",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphamissense",
    "verification": "verified",
    "note": null
   },
   "note": "The prediction catalogue is open including commercial use and the code is open; no weights release or licence text was retrieved this round"
  },
  "first_release": {
   "date": "2023-09-19",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal paper and developer blog on the same day",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1126/science.adg7492",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-09-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal paper published",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1126/science.adg7492",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": {
    "date": "2023-09-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal online",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1126/science.adg7492",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2023-09-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Code and prediction catalogue released with the paper",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/google-deepmind/alphamissense",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2023-09-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer project page and blog",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://deepmind.google/research/projects/alphamissense/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2024-03-13",
   "precision": "day",
   "year_hint": null,
   "basis": "Last repository commit: predictions-database licence update",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/google-deepmind/alphamissense",
   "verification": "verified",
   "description": "The latest sourced event for this model is the 2024-03 predictions-database licence update, after which the repository was archived. Note (not an update to this model): a 2026-09-08 variant-impact score is a downstream product of a different line and is sourced only second-hand",
   "note": "The site recorded 2026-09-08 as an update to this model; corrected"
  },
  "status": "maintained",
  "status_note": "No substantive event for the model itself since 2024-03; repository archived",
  "what_it_does": "Starts from a co-folding protein structure model and fine-tunes on weak labels from variants that are common or absent in human and primate populations, scoring every possible single amino-acid substitution and publishing most of them as a precomputed catalogue for clinical and research lookup. Like popEVE it predicts missense effects, but it emphasises a proteome-wide catalogue while the other emphasises cross-gene calibration and rare-disease cohorts; sibling lines are AlphaGenome for genomics and AlphaProteo for protein design.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Evolutionary information from protein sequence and structure plus allele frequencies from public population resources",
    "source_url": "https://doi.org/10.1126/science.adg7492"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1126/science.adg7492"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No own wet lab; model selection used a clinical variant database and the developers' own figure note concedes such methods may be overestimated on that benchmark",
    "source_url": "https://deepmind.google/research/projects/alphamissense/"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-11-clinvar",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.1126/science.adg7492",
    "confidence": "high",
    "second_hand": false,
    "note": "A developer-built held-out clinical variant test set (not an official split) plus a gene-level aggregate; the protocol is the developers' own"
   },
   {
    "benchmark_id": "bench-13-cagi",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11976771/",
    "confidence": "medium",
    "second_hand": false,
    "note": "A third-party post-hoc evaluation, not participation: the source calls it the best single (non-meta) predictor, tied with another non-meta predictor and behind two meta predictors, and gives no overall rank"
   }
  ],
  "other_evaluations": [
   {
    "label": "Deep mutational scanning correlation",
    "category": "public-benchmark-not-on-site",
    "detail": "The developer project page reports correlation on DMS data for about two dozen proteins, outside the 44",
    "source_type": "self"
   },
   {
    "label": "Third-party variant-model comparison (preprint, abstract level)",
    "category": "comparison-vs-baseline-only",
    "detail": "Another variant model's preprint uses it as a baseline and concedes partial tuning on a clinical variant database; abstract level only",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: combining a structure model with population-frequency weak labels yields directly queryable pathogenicity predictions for proteome-wide missense substitutions and outperforms earlier evolutionary predictors.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A third-party post-hoc evaluation exists (see the CAGI entry): it is listed as the best single non-meta predictor but behind meta predictors, and it was not a participant.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The developer is the AI research unit of a listed US technology group (headquartered in London), funded internally with no separate financing; the prediction catalogue is free to use including commercially.",
   "funding": null,
   "products_partnerships": "The model derives from the group's co-folding structure series; a variant-impact score released on 2026-09-08 by a separate genomics product line is reported to combine this model's outputs, but only in second-hand sources and is not treated as an event for this model.",
   "related_models": [
    "model-alphafold3",
    "model-alphagenome",
    "model-alphaproteo",
    "model-popeve"
   ]
  },
  "reading_notes": [
   "The self-reported clinical-variant figures come from a developer-built held-out set; same-named protocols in different papers are not directly comparable.",
   "High discrimination is not clinical usability; evidence strength depends on the likelihood ratio at the chosen threshold.",
   "Nearly every later variant model cites it as a baseline, yet it does not appear in the official ProteinGym summary tables, so \"cited as a baseline\" must not be read as \"on the board\".",
   "The leakage concern stems from model selection on a clinical variant database, not from training on it directly."
  ],
  "caveats": [
   "The CAGI entry is a post-hoc evaluation, not participation, and the source gives no overall rank.",
   "The held-out clinical variant set is developer-built, not an official split.",
   "No substantive event for the model itself since 2024-03; the repository is archived.",
   "The 2026 downstream score belongs to another product line and is sourced only second-hand."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1126/science.adg7492",
    "title": "Journal paper 2023",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://deepmind.google/research/projects/alphamissense/",
    "title": "Developer project page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/google-deepmind/alphamissense",
    "title": "Code and predictions repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11976771/",
    "title": "Third-party post-hoc evaluation paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://siliconangle.com/2026/09/08/google-deepminds-alphagenome-at",
    "title": "Downstream product line coverage (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported[bench-13-cagi]",
    "was": "Ranked first among 60 predictors in CAGI6",
    "now": "Post-hoc evaluation, not a participant; the paper calls it the best single (non-meta) predictor, tied with another non-meta predictor and behind two meta-predictors, with no overall ranking given",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11976771/",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2026-09-08, product update",
    "now": "2024-03-13, no later sourced event found (the 2026 event belongs to another product line and is second-hand only)",
    "source_url": "https://github.com/google-deepmind/alphamissense",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Candidate pairing ProteinGym",
    "now": "Not a pairing: no entry for this model in either the official clinical or the DMS zero-shot summary tables",
    "source_url": "https://doi.org/10.1126/science.adg7492",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "background.products_partnerships",
   "release_milestones.code_release",
   "open_status.weights",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-alphamissense/",
   "en": "/bio/benchmark/models/alphamissense/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1,
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-alphaproteo",
  "kind": "model",
  "short_slug": "alphaproteo",
  "names": {
   "display": "AlphaProteo",
   "full": null,
   "full_null_reason": "Sources use AlphaProteo only; no expanded name",
   "aliases": [
    "AlphaProteo"
   ],
   "ambiguous_aliases": [
    "AlphaFold"
   ]
  },
  "family": null,
  "family_null_reason": "One technical-report version only; no family members",
  "domain": "protein",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Closed de novo protein binder design system; seven targets reported successful and an eighth failed",
  "org": {
   "names": [
    "Google DeepMind",
    "Francis Crick Institute"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Design (de novo binder)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Closed: no code and no weights; still no release plan as of 2026-04 (second-hand)"
  },
  "first_release": {
   "date": "2024-09-05",
   "precision": "day",
   "event_kind": "official-announcement",
   "basis": "Developer blog post (the technical report on arXiv, 2024-09-12, is listed separately)",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2409.08022",
   "verification": "partial",
   "note": "The blog date comes from the existing note and search results; the page was not fetched this round"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-09-05",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Developer blog post (the technical report on arXiv, 2024-09-12, is listed separately)",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2409.08022",
    "verification": "partial",
    "note": "The blog date comes from the existing note and search results; the page was not fetched this round"
   },
   "paper_v1": {
    "date": "2024-09-12",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Technical report arXiv v1 (date carried from the existing note)",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2409.08022",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-09-05",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Developer blog post (the technical report on arXiv, 2024-09-12, is listed separately)",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2409.08022",
    "verification": "partial",
    "note": "The blog date comes from the existing note and search results; the page was not fetched this round"
   }
  },
  "latest_update": {
   "date": "2024-09-05",
   "precision": "day",
   "event_kind": "none-since-release",
   "basis": "Only the release event is sourced; no new version, weights or successor found in 2025-2026",
   "description": "No later sourced event found (as of 2026-11-16); whether the arXiv v1 report (2024-09-12) counts as an update is a policy question and is unverified.",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2409.08022",
   "verification": "partial",
   "note": "Must never be phrased as 'no updates since'"
  },
  "status": "maintained",
  "status_note": "Closed; only the 2024-09 release event is sourced",
  "what_it_does": "Given a target structure and optional hotspot residues it generates and filters candidate binder sequences, aimed at one design round plus a small wet-lab test. It was among the first big-lab design models to publish wet-lab data, and Latent-X/X2, BoltzGen and BindCraft all use it as a comparator, but it is closed and cannot be reproduced externally.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Distilled PDB and predicted-structure data; only the source databases are named, with no size, cutoff or filtering rules",
    "source_url": "https://arxiv.org/abs/2409.08022"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "The report does not mention private or licensed data",
    "source_url": "https://arxiv.org/abs/2409.08022"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "Whether own experimental data entered training is not disclosed",
    "source_url": "https://arxiv.org/abs/2409.08022"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Yes: the developer's wet lab ran binding tests on eight targets, with partner functional assays",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Eight-target binder campaign (self-reported)",
    "modality": "protein binder",
    "targets": 8,
    "designs_tested": null,
    "hits": 7,
    "denominator_basis": "Eight targets attempted (BHRF1, SARS-CoV-2 RBD, IL-7Ra, PD-L1, TrkA, IL-17A, VEGF-A and TNFa) in one round of medium-throughput screening",
    "success_rate": "binders on 7 of 8 targets (TNFa is the failed eighth target)",
    "affinity_best": "developers report 3- to 300-fold better affinities than prior methods (self-reported)",
    "version": "blog plus technical report",
    "source_type": "self",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "date": {
     "date": "2024-09-05",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Developer blog post (the technical report on arXiv, 2024-09-12, is listed separately)",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2409.08022",
     "verification": "partial",
     "note": "The blog date comes from the existing note and search results; the page was not fetched this round"
    },
    "verification": "verified",
    "note": "Self-reported with no independent replication; TNFa must be described as the failed eighth target and never merged into the seven successes"
   },
   {
    "campaign": "BHRF1 per-target binding rate (unverified)",
    "modality": "protein binder",
    "targets": 1,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "The body text and a figure caption of the same developer source give different per-target rates, so no figures are given (unverified)",
    "success_rate": null,
    "affinity_best": null,
    "version": "blog",
    "source_type": "self",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "date": {
     "date": "2024-09-05",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Developer blog post (the technical report on arXiv, 2024-09-12, is listed separately)",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2409.08022",
     "verification": "partial",
     "note": "The blog date comes from the existing note and search results; the page was not fetched this round"
    },
    "verification": "partial",
    "note": "The body text is the authoritative reading, but the source is internally inconsistent, so this record marks it partial and quotes no value"
   },
   {
    "campaign": "Partner-lab functional assays",
    "modality": "protein binder",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Qualitative, no counts: some RBD binders blocked infection in cells",
    "success_rate": null,
    "affinity_best": null,
    "version": "blog",
    "source_type": "partner-lab",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "date": {
     "date": "2024-09-05",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Developer blog post (the technical report on arXiv, 2024-09-12, is listed separately)",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2409.08022",
     "verification": "partial",
     "note": "The blog date comes from the existing note and search results; the page was not fetched this round"
    },
    "verification": "verified",
    "note": "Partner-lab executed; not an independent replication"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Seven-target binder wet-lab campaign",
    "category": "wet-lab-validation",
    "detail": "One round of medium-throughput screening with self-reported better affinity and success than prior methods; VEGF-A succeeded and TNFa failed",
    "source_type": "self"
   },
   {
    "label": "Success-rate comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "Per-target success compared with the best prior methods; the BHRF1 value is internally inconsistent and unverified",
    "source_type": "self"
   },
   {
    "label": "Follow-on functional tests",
    "category": "case-study",
    "detail": "The release blog describes VEGF-A and SARS-CoV-2 RBD binders tested in partner-lab assays",
    "source_type": "partner"
   }
  ],
  "claimed_advantage": "Developers report that designs against seven targets, including the difficult VEGF-A, bound with 3- to 300-fold better affinity and higher success than the best prior methods after a single screening round, without iterative optimisation (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Closed and not externally runnable; later competitors only cite its numbers as comparison",
  "background": {
   "org_profile": "Google DeepMind is Alphabet's AI research unit (two teams merged in 2023), based in London and funded internally rather than externally; its protein design team previously built the structure-prediction series. A UK research institute contributed the wet-lab work.",
   "funding": null,
   "funding_null_reason": "Internally funded by Alphabet; no external funding rounds",
   "products_partnerships": "Not commercialised; the developer says it is also exploring applications with a sister drug-design company.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3",
    "model-alphamissense",
    "model-latent-x2",
    "model-bindcraft"
   ]
  },
  "reading_notes": [
   "The eight targets were self-chosen and the figures are within-target hit rates, not comparable to open-competition rates.",
   "The TNFa failure shows it is no universal solution; the seven successes must be read together with that eighth failure.",
   "Being closed means no third party can re-test it, so every figure stays self-reported.",
   "Only source databases are disclosed, with no scale or cutoff, so overlap with the evaluation targets cannot be checked."
  ],
  "caveats": [
   "Zero pairings: it reports no score on any site benchmark.",
   "No entry or submission record in the Adaptyv competition / BenchBB; the model is closed and cannot be run externally.",
   "The BHRF1 per-target rate is internally inconsistent within one source and is treated as unverified.",
   "Training data is disclosed only at database level and must not be written up as 'public data only'."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2409.08022",
    "title": "AlphaProteo technical report (arXiv 2409.08022)",
    "kind": "preprint",
    "date": {
     "date": "2024-09-12",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "arXiv v1",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2409.08022",
     "verification": "partial",
     "note": null
    }
   },
   {
    "url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "title": "Developer release blog",
    "kind": "company-blog",
    "date": {
     "date": "2024-09-05",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Blog published",
     "timestamp_utc": null,
     "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
     "verification": "partial",
     "note": null
    }
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "TNFa was presented as one of the seven successful targets",
    "now": "TNFa is the failed eighth target and is listed apart from the seven successes",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "The BHRF1 per-target rate was published as a single value",
    "now": "Body text and figure caption disagree, so it is marked partial and no figure is given",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "The latest update was phrased as there being no updates since release",
    "now": "Changed to 'no later sourced event found (as of 2026-11-16)' with event kind none-since-release",
    "source_url": "https://arxiv.org/abs/2409.08022",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "first_release",
   "wetlab_validation[1]",
   "training_data.b_private_licensed",
   "training_data.c_own_wetlab_for_training"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-alphaproteo/",
   "en": "/bio/benchmark/models/alphaproteo/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-amplify",
  "kind": "model",
  "short_slug": "amplify",
  "names": {
   "display": "AMPLIFY",
   "full": null,
   "full_null_reason": "The name is a model codename with no expanded form in the paper",
   "aliases": [
    "AMPLIFY",
    "AMPLIFY 350M",
    "AMPLIFY 120M"
   ],
   "ambiguous_aliases": [
    "Amplify"
   ]
  },
  "family": null,
  "family_null_reason": "120M and 350M are two sizes of one model, not separate members",
  "domain": "protein",
  "domains_secondary": [],
  "one_liner": "Small protein language models arguing data quality matters more than scale",
  "org": {
   "names": [
    "Mila",
    "Polytechnique Montréal",
    "Amgen"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": "A joint project of an academic institute and a university lab with a large biopharma company's research arm",
  "country": [
   "CA",
   "US"
  ],
  "model_class": "Protein language model (small scale)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": null,
   "license_changed": null,
   "note": "Code, data and weights are all public and the model card gives MIT for the weights; another page says Apache 2.0, so the conflicting code terms are left empty"
  },
  "first_release": {
   "date": "2024-09-23",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "Preprint v1 posted; the 2024-09-19 first commits and 2024-09-30 weights do not count as first release",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2024.09.23.614603",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-09-23",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2024.09.23.614603",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-09-23",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2024.09.23.614603",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-09-19",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "First repository commits",
    "timestamp_utc": null,
    "source_url": "https://github.com/chandar-lab/AMPLIFY/commits/main",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2024-09-30",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights published to a public model hub",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/chandar-lab/AMPLIFY_350M/commits/main",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-01-05",
   "precision": "day",
   "event_kind": "paper-revision",
   "basis": "Preprint v2 revision",
   "description": "Paper revised to v2 (v1 was 2024-09-23); no journal version, and the repository and model hub carry only minor changes with the last default-branch commit on 2025-07-17",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.23.614603",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only a paper revision and minor code changes, no new generation",
  "what_it_does": "Uses a modern architecture, cleaned and debiased training data and an efficient codebase to train protein language models far smaller than their contemporaries, to test whether bigger is always better. It is a research baseline rather than a product, positioned against the scale-first route of ESM3 and ESM C.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Three public databases combined into one public dataset: a sequence database, an antibody database and a structural-classification database",
    "source_url": "https://huggingface.co/datasets/chandar-lab/UR100P"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data reported",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Held-out UniProt pseudo-perplexity",
    "category": "comparison-vs-baseline-only",
    "detail": "Held-out likelihood against existing protein language models at several scales, claiming the small model is competitive with the largest baseline",
    "source_type": "self"
   },
   {
    "label": "Masked-residue sequence recovery",
    "category": "own-task-set",
    "detail": "Recovery on the human proteome grouped by conservation plus a long-context extension, including the authors' own retrained controls",
    "source_type": "self"
   },
   {
    "label": "Unsupervised contact prediction",
    "category": "public-benchmark-not-on-site",
    "detail": "Attention-map contact prediction with logistic regression and random forest: the baseline is better with logistic regression and the gap shrinks with random forest",
    "source_type": "self"
   },
   {
    "label": "Disorder discrimination",
    "category": "public-benchmark-not-on-site",
    "detail": "Discrimination of a disorder database against hypothetical proteins, with embedding similarity compared to structure-prediction confidence",
    "source_type": "self"
   },
   {
    "label": "Inference throughput and design pipeline",
    "category": "comparison-vs-baseline-only",
    "detail": "Proteins per second on one GPU, claimed more than an order of magnitude faster than the largest baseline, plus sequence recovery in a folding-and-inverse-folding pipeline",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that a 350M model matches or arguably beats a baseline with 43x more parameters on pseudo-perplexity, with about 17x less training compute and much faster inference, attributing this to data quality.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "A row-by-row check of the official board found no entry, and no third-party evaluation was found",
  "background": {
   "org_profile": "An academic-industry collaboration with no separate company: an AI research institute in Montreal, Canada and a lab at a local engineering university, together with the research arm of a large US biopharma company headquartered in California.",
   "funding": null,
   "funding_null_reason": "The paper does not itemise funding; nothing inferred",
   "products_partnerships": "Releases 120M and 350M sizes together with the training dataset and is included in a GPU vendor's bio-model framework. Its research question targets the scale-first route represented by ESM3 and ESM C.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-esm3",
    "model-esm-c",
    "model-saprot"
   ]
  },
  "reading_notes": [
   "Its contribution is methodological; it has no entry on the official zero-shot board and cannot enter board comparisons.",
   "The \"beats the largest baseline\" claim is limited to \"some tasks\" with no public task list or figures.",
   "Most self-evaluations are own-built or off-site benchmarks, not comparable with the 44 on this site.",
   "Fully open, dataset included, so it is cheap to reproduce and useful as a control baseline."
  ],
  "caveats": [
   "Zero pairings: under this round's relation rules it has no established pairing with the 44 site benchmarks.",
   "The task list behind the claimed advantage is not public and cannot be checked (unverified).",
   "Licence statements conflict: MIT for the weights, Apache 2.0 elsewhere.",
   "No wet lab and no third-party evaluation."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2024.09.23.614603",
    "title": "AMPLIFY preprint (v1 2024-09, v2 2026-01)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/chandar-lab/AMPLIFY_120M",
    "title": "Weights and model card (120M)",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/datasets/chandar-lab/UR100P",
    "title": "Public training dataset",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/chandar-lab/AMPLIFY/commits/main",
    "title": "Official repository commit history",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://mila.quebec/en/insight/is-bigger-always-better-democratizing-ai-protein-discovery",
    "title": "Institute explainer page with the developer framing",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://docs.nvidia.com/bionemo-framework/2.6.3/models/amplify",
    "title": "Third-party framework documentation",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.23.614603",
    "title": "Preprint version metadata",
    "kind": "archive",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "ProteinGym official summary table, checked row by row with no entry",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.event_kind",
    "was": "Recorded as a new version",
    "now": "Changed to a paper revision: 2026-01-05 is preprint v2, not a new model version, and no journal version exists",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.23.614603",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "Said only \"v2 revision\" plus one code change",
    "now": "Adds the v1 and v2 dates, states there is no journal version and that code changes are minor with the last commit on 2025-07-17",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.23.614603",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "claimed_advantage",
   "open_status.code_license",
   "release_milestones.code_release",
   "release_milestones.weights_release"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-amplify/",
   "en": "/bio/benchmark/models/amplify/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-aqaffinity",
  "kind": "model",
  "short_slug": "aqaffinity",
  "names": {
   "display": "AQAffinity",
   "full": null,
   "aliases": [
    "AQAffinity"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single version with no family members",
  "domain": "small-molecule",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Open protein-ligand affinity model adding an affinity head to an open co-folding backbone",
  "org": {
   "names": [
    "SandboxAQ"
   ],
   "global_entity_ids": []
  },
  "org_type": "platform-company",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding / affinity prediction (affinity head)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Weights plus training and inference code are open, positioned as a licence-clean commercially usable affinity baseline; the weights host page requires click-through approval"
  },
  "first_release": {
   "date": "2026-01-20",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer blog post date (the weights repository was created 2025-12-05, but repository creation is inadmissible as a first release)",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-01-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog post",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2026-01-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Training and inference code released with the blog post",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2026-01-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights public (the host page is gated with automatic approval)",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/SandboxAQ/AQAffinity",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-01-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog post",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-02-10",
   "precision": "day",
   "year_hint": null,
   "basis": "Last modified time of the weights repository",
   "event_kind": "patch",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/api/models/SandboxAQ/AQAffinity",
   "verification": "unverified",
   "description": "The latest sourced event is the weights repository's last modification; the commit history is gated so the nature of the change cannot be verified, and no new weights or version were found",
   "note": "Must not be recorded as maintenance only: for a weights repository that reading is unsupported and the field displays as unverified"
  },
  "status": "active",
  "status_note": "Weights and code released within 12 months",
  "what_it_does": "Attaches a reimplemented affinity head to the open co-folding backbone OpenFold3 to predict protein-ligand affinity; it differs from Boltz-2 mainly in backbone and licence rather than modelling idea, while TerraBind in the same direction is closed and relies on internal data.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Three public activity and compound databases",
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "A licensed commercial activity database was used only for evaluation, not training",
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No wet-lab validation reported",
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
    "confidence": "medium",
    "second_hand": false,
    "note": "Post-hoc use of two affinity subsets from a released round, shown only as a figure with no values, and one subset closely resembles the training targets"
   }
  ],
  "other_evaluations": [
   {
    "label": "Out-of-distribution assay comparison on a licensed database",
    "category": "comparison-vs-baseline-only",
    "detail": "The repository itself states it is clearly weaker than the model it reimplements on two out-of-distribution assays",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a licence-clean, commercially usable open structure-to-affinity baseline shipped with training code, rather than an accuracy breakthrough.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation found; all existing comparisons come from developer materials",
  "background": {
   "org_profile": "An AI and quantum technology company spun out in 2022-03 from a quantum team formed inside a US technology group in 2016; headquartered in California (about 55 staff at spin-out, with press releases of different years giving different cities).",
   "funding": "A $500M Series D in 2022-03; a Series E in 2025-04 totalling about $450M with an additional $150M bringing in two large technology companies at a valuation of about $5.75B, roughly $950M raised in total (all from wire-service reporting with slightly varying framings).",
   "products_partnerships": "Its drug-discovery unit launched 2023-06 with two large pharma companies named as customers; on 2025-06-18 it released a synthetic protein-ligand structure dataset with affinity labels (licence wording differs across pages); in 2023-11 it announced an AI-simulation collaboration with a chipmaker; it participates in an open co-folding consortium whose backbone carries this affinity head.",
   "related_models": [
    "model-openfold3",
    "model-boltz-2",
    "model-terrabind",
    "model-nesso-1"
   ]
  },
  "reading_notes": [
   "The value lies in the open licence and training code, not in leading accuracy.",
   "In the developers' own comparison it trails the model it reimplements, so it should not be treated as state of the art.",
   "One CASP subset closely resembles the training targets, so the figure cannot stand for generalisation, and it gives no values.",
   "The nature of the latest change is unverifiable (the commit history is gated), so do not record it as maintenance only."
  ],
  "caveats": [
   "The nature of the latest update is unverified; the repository commit history is not visible.",
   "No verified site benchmark value exists; the only entry is retrospective dataset use.",
   "No wet-lab validation.",
   "No third-party evaluation yet."
  ],
  "sources": [
   {
    "url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
    "title": "Official release blog post",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://huggingface.co/SandboxAQ/AQAffinity",
    "title": "Weights and model card",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://huggingface.co/api/models/SandboxAQ/AQAffinity",
    "title": "Weights repository metadata API",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.sandboxaq.com/press/sandboxaq-unveils-sair-structurally-augmented-ic50-repository-a-novel-open-dataset-of-protein-ligand-structures-labelled-by-binding-affinities",
    "title": "Dataset release press release",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update_kind",
    "was": "Maintenance only (code-maintenance-only)",
    "now": "Changed to a patch-type event and marked unverified: the weights repository commit history requires approval, so the nature of the change has no basis",
    "source_url": "https://huggingface.co/api/models/SandboxAQ/AQAffinity",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "The basis included the weights repository creation date (2025-12-05)",
    "now": "2026-01-20, based on the official blog post",
    "source_url": "https://www.sandboxaq.com/post/aqaffinity-open-source-structure-to-affinity-built-on-openfold3-by-sandboxaq",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "release_milestones.code_release",
   "release_milestones.weights_release",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-aqaffinity/",
   "en": "/bio/benchmark/models/aqaffinity/"
  },
  "names_full_null_reason": "Official materials give no expanded name",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-beacon",
  "kind": "model",
  "short_slug": "beacon",
  "names": {
   "display": "Beacon (Inductive Bio)",
   "full": null,
   "aliases": [
    "Beacon-2",
    "Inductive Bio Beacon"
   ],
   "ambiguous_aliases": [
    "Beacon"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Beacon",
     "id_or_null": null,
     "first_release": {
      "date": "2025-05",
      "precision": "month",
      "year_hint": null,
      "basis": "Press coverage of the company leaving stealth mentions the Beacon platform (model release date not given)",
      "event_kind": "product-update",
      "timestamp_utc": null,
      "source_url": "https://www.statnews.com/2025/05/07/flatiron-health-veterans-raise-25-million-ai-tool-drug-toxicity-inductive-bio/",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "Beacon-2",
     "id_or_null": null,
     "first_release": {
      "date": "2026-09-17",
      "precision": "day",
      "year_hint": null,
      "basis": "Launch date in secondary reports (original press release not retrieved)",
      "event_kind": "successor-model",
      "timestamp_utc": null,
      "source_url": "https://runtimewire.com/article/inductive-beacon-2-human-dose-drug-discovery",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "Beacon is the ADMET model; Beacon-2 adds human dose estimation, and the challenge wins belong to Beacon, not Beacon-2"
  },
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "Inductive Bio's closed ADMET/PK predictor trained on a pre-competitive pharma data consortium",
  "org": {
   "names": [
    "Inductive Bio"
   ],
   "global_entity_ids": []
  },
  "org_type": "platform-company",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (ADMET / PK)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Closed; provided to partners through the company platform"
  },
  "first_release": {
   "date": "2025-05",
   "precision": "month",
   "year_hint": null,
   "basis": "Press coverage of the company with its Beacon platform (model release date not given; month approximate)",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.statnews.com/2025/05/07/flatiron-health-veterans-raise-25-million-ai-tool-drug-toxicity-inductive-bio/",
   "verification": "partial",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-05",
    "precision": "month",
    "year_hint": null,
    "basis": "Press coverage",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.statnews.com/2025/05/07/flatiron-health-veterans-raise-25-million-ai-tool-drug-toxicity-inductive-bio/",
    "verification": "partial",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-05",
    "precision": "month",
    "year_hint": null,
    "basis": "Beacon described in Series A coverage",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://alleywatch.com/2025/05/inductive-bio-ai-drug-discovery-small-molecule-optimization-admet-collaboration-platform-josh-haimson/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09-17",
   "precision": "day",
   "year_hint": null,
   "basis": "Beacon-2 launch (secondary report)",
   "event_kind": "successor-model",
   "timestamp_utc": null,
   "source_url": "https://runtimewire.com/article/inductive-beacon-2-human-dose-drug-discovery",
   "verification": "partial",
   "description": "Beacon-2 adds human efficacious-dose prediction; a medicinal-chemistry agent, Indy, was introduced days earlier",
   "note": "The original press release was not opened; date from secondary reports"
  },
  "status": "product-only",
  "status_note": "Closed commercial product with successor Beacon-2",
  "what_it_does": "Predicts ADMET and pharmacokinetic properties before a compound is made, fine-tuned for each partner's chemical space; Beacon-2 adds human dose estimates. It shares the ADMET route with ADMET-AI and MiniMol, but its training data come from a consortium pooling IP-protected data from several companies.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public data used; libraries not named",
    "source_url": "https://runtimewire.com/article/inductive-beacon-2-human-dose-drug-discovery"
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "A pre-competitive consortium pooling anonymised partner data; compound and assay counts unpublished",
    "source_url": "https://alleywatch.com/2025/05/inductive-bio-ai-drug-discovery-small-molecule-optimization-admet-collaboration-platform-josh-haimson/"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "No statement of own experimental data",
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Blind challenges are external tests, not own wet lab",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-27-polaris",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://openadmet.ghost.io/lessons-learned-from-the-openadmet-expansionrx-blind-challenge/",
    "confidence": "high",
    "second_hand": false,
    "note": "The company press release claims first place in three blind challenges; the boards are pseudonymous and cannot be matched to a model version, and ExpansionRx and PXR were run by OpenADMET outside the Polaris hub, so under this round's scope they fall outside the benchmark; not counted"
   }
  ],
  "other_evaluations": [
   {
    "label": "OpenADMET blind challenges",
    "category": "public-benchmark-not-on-site",
    "detail": "The press release claims first place in antiviral, ExpansionRx and PXR challenges; only one is echoed by a third-party write-up and no official final board was read",
    "source_type": "self"
   },
   {
    "label": "Beacon-2 dose estimation",
    "category": "own-task-set",
    "detail": "Dose prediction on ExpansionRx public compounds (secondary reports); sources give inconsistent figures",
    "source_type": "self"
   },
   {
    "label": "Partner fine-tuning",
    "category": "user-or-partner-report",
    "detail": "Fine-tuned per partner on consortium data; scale not disclosed",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: repeated first places in blind ADMET challenges, with consortium data and partner fine-tuning capturing subtle structural signals; Beacon-2 adds human dose estimates.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The challenges are organiser-run, but only press releases and secondary reports were read; the boards are pseudonymous and no matching entry was verified",
  "background": {
   "org_profile": "Inductive Bio is based in New York and does not develop its own drugs; it commercialises a pre-competitive consortium where companies contribute anonymised data under IP protection and clients pay software and collaboration fees. It left stealth in 2023-12.",
   "funding": "About $4.3M seed disclosed at launch (a16z Bio + Health and Lux Capital); a $25M Series A led by Obvious Ventures announced in 2025-05.",
   "products_partnerships": "Products are the Compass platform, the Beacon models and the chemistry assistant Indy; Denali Therapeutics is an early partner; with Ginkgo Datapoints and Tangible Scientific it forms an assay, modelling and compound-management workflow.",
   "related_models": [
    "model-admet-ai",
    "model-minimol"
   ]
  },
  "reading_notes": [
   "First places suggest strong results on those datasets, but the evidence is company press releases and the boards are pseudonymous.",
   "Consortium data size is undisclosed, so it is unclear whether the edge comes from the model or the data.",
   "\"Consecutive wins\" is a point-in-time claim and needs a date when cited.",
   "The two second-hand dose-estimation figures for Beacon-2 must not be mixed."
  ],
  "caveats": [
   "The Polaris pairing is not established: identity cannot be verified and two challenges were outside the Polaris hub.",
   "Beacon-2's launch date and description come from secondary reports.",
   "The first-release month is approximate; no public release date exists for the model itself."
  ],
  "sources": [
   {
    "url": "https://www.streetinsider.com/PRNewswire/Inductive+Bio+Wins+3rd+Consecutive+AI+Drug+Discovery+Benchmark/26765853.html",
    "title": "Company press release (PXR challenge)",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://runtimewire.com/article/inductive-beacon-2-human-dose-drug-discovery",
    "title": "Beacon-2 coverage",
    "kind": "second-hand",
    "date": null
   },
   {
    "url": "https://www.statnews.com/2025/05/07/flatiron-health-veterans-raise-25-million-ai-tool-drug-toxicity-inductive-bio/",
    "title": "Series A coverage",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://alleywatch.com/2025/05/inductive-bio-ai-drug-discovery-small-molecule-optimization-admet-collaboration-platform-josh-haimson/",
    "title": "Company interview coverage",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://openadmet.ghost.io/lessons-learned-from-the-openadmet-expansionrx-blind-challenge/",
    "title": "OpenADMET challenge lessons write-up",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://www.biospace.com/inductive-bio-emerges-from-stealth-unveiling-an-ml-platform-to-accelerate-compound-optimization-in-drug-discovery",
    "title": "Stealth exit coverage",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "First in three Polaris blind challenges (self-reported)",
    "now": "Unconfirmed and not counted: the board is anonymous, and two of the events are not on the Polaris platform",
    "source_url": "https://openadmet.ghost.io/lessons-learned-from-the-openadmet-expansionrx-blind-challenge/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "latest_update",
   "release_milestones.first_public_evidence",
   "training_data.c_own_wetlab_for_training",
   "benchmarks_reported[0]"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-beacon/",
   "en": "/bio/benchmark/models/beacon/"
  },
  "names_full_null_reason": "A product name with no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-bindcraft",
  "kind": "model",
  "short_slug": "bindcraft",
  "names": {
   "display": "BindCraft",
   "full": null,
   "full_null_reason": "Sources use BindCraft only; no expanded name",
   "aliases": [
    "BindCraft"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single pipeline with version iterations; no family members",
  "domain": "protein",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Open binder design pipeline that backpropagates through a co-folding network, with no extra training",
  "org": {
   "names": [
    "EPFL Laboratory of Protein Design and Immunoengineering",
    "MIT"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CH"
  ],
  "model_class": "Design (hallucination via backpropagation)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code is open; it has no weights of its own and depends on a third-party co-folding model's weights and licence"
  },
  "first_release": {
   "date": "2024-10-01",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 (later v2 2024-12-07, v3 2025-04-25)",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.30.615802",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-10-01",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1 (later v2 2024-12-07, v3 2025-04-25)",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.30.615802",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-10-01",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1 (later v2 2024-12-07, v3 2025-04-25)",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.30.615802",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-08-27",
    "precision": "day",
    "event_kind": "journal-online",
    "basis": "Journal version online (Crossref; print issue 2025-10-09)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41586-025-09429-6",
    "verification": "verified",
    "note": "Listed as its own milestone and never merged with the 2026 patch"
   },
   "code_release": {
    "date": null,
    "precision": null,
    "event_kind": "code-release",
    "basis": "The repository is public but its first release date was not retrieved this round",
    "timestamp_utc": null,
    "source_url": "https://github.com/martinpacesa/BindCraft",
    "verification": "unverified",
    "note": "Unverified: only continuing commits through 2026-09-21 are known"
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-04-17",
   "precision": "day",
   "event_kind": "patch",
   "basis": "Tag v1.5.3 (small fixes)",
   "description": "v1.5.3 small fixes; the journal version online 2025-08-27 is a separate milestone; commits to 2026-09-21 are install and notebook maintenance.",
   "timestamp_utc": null,
   "source_url": "https://github.com/martinpacesa/BindCraft/releases",
   "verification": "verified",
   "note": "The earlier record treated this date as the journal publication; corrected to a patch"
  },
  "status": "active",
  "status_note": "Tagged release and ongoing commits within 12 months",
  "what_it_does": "It folds target and binder together in a co-folding network and iteratively edits the binder sequence to maximise predicted confidence, allowing some induced fit; there is no generative model and no training. Compared with RFdiffusion2/3, BoltzGen and Germinal it is the lightest and easiest to reproduce.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "No training of its own; it uses a third-party co-folding model's off-the-shelf weights, trained on PDB and other public data",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No own training data",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Yes: the authors ran the wet-lab validation themselves",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Experimental binder success across 12 targets (self-reported)",
    "modality": "mini-binder",
    "targets": 12,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Twelve targets; the rate is the share of tested designs that bound, and per-target denominators are very small",
    "success_rate": "10-100% per target, average 46.3% (journal body text)",
    "affinity_best": null,
    "version": "journal version",
    "source_type": "self",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6",
    "date": {
     "date": "2025-08-27",
     "precision": "day",
     "event_kind": "journal-online",
     "basis": "Journal version online (Crossref; print issue 2025-10-09)",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1038/s41586-025-09429-6",
     "verification": "verified",
     "note": "Listed as its own milestone and never merged with the 2026 patch"
    },
    "verification": "verified",
    "note": "The 46.3% figure is stated directly in the journal body text, so the earlier 'second-hand summary' label is dropped; the authors ran the experiments themselves, so this is not independent replication"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-28-adaptyv",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "Nipah and EGFR rounds",
    "source_url": "https://adaptyvbio.substack.com/p/protein-design-competition-results",
    "confidence": "medium",
    "verification": "partial",
    "note": "The authors entered the Nipah round and the tool appears in an EGFR round top-10; the organisers publish no per-tool hit rates, so the record is second-hand"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own 12-target wet-lab set",
    "category": "wet-lab-validation",
    "detail": "Self-chosen targets with the share of tested designs that bound; per-target denominators are very small (self-reported)",
    "source_type": "self"
   },
   {
    "label": "Application demonstrations",
    "category": "case-study",
    "detail": "Binders blocking a pollen allergen and peptides improving a gene-therapy vector; the authors say they are extending to peptides",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that with no extra training most targets yield binders in one design round, with about 46.3% of tested designs binding (journal body text, self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The competition publishes no per-tool hit rates, so performance there rests on entrants' own accounts (second-hand)",
  "background": {
   "org_profile": "An academic project of a protein design and immunoengineering laboratory in the engineering school of a Swiss institute of technology, with no company entity; the university's news release describes collaboration with a US university and researchers from Switzerland, the US and the Netherlands.",
   "funding": null,
   "funding_null_reason": "Grants are listed only in the journal acknowledgements, not in the news release",
   "products_partnerships": null,
   "products_partnerships_null_reason": "Academic project; no product and no pharma partnership",
   "related_models": [
    "model-rfdiffusion2-3",
    "model-boltzgen",
    "model-germinal",
    "model-mosaic"
   ]
  },
  "reading_notes": [
   "The 10-100% span is wide and its average depends on target choice, so it is not an expected rate for an arbitrary target.",
   "Optimising a co-folding confidence score as the design objective disqualifies that score as an independent filter, a known issue for this family.",
   "The Adaptyv competition does not report per tool, so performance there rests on entrants' own accounts.",
   "Being open and training-free is why it is so widely used as a baseline."
  ],
  "caveats": [
   "The success rate is the share of tested designs that bound, with very small per-target denominators.",
   "It reports no site benchmark score; its only pairing is a competition entry record (second-hand).",
   "It has no weights of its own, so design quality inherits the biases of the co-folding model it depends on.",
   "The repository's first release date is unverified."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41586-025-09429-6",
    "title": "Journal paper DOI",
    "kind": "journal",
    "date": {
     "date": "2025-08-27",
     "precision": "day",
     "event_kind": "journal-online",
     "basis": "Journal version online (Crossref; print issue 2025-10-09)",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1038/s41586-025-09429-6",
     "verification": "verified",
     "note": "Listed as its own milestone and never merged with the 2026 patch"
    }
   },
   {
    "url": "https://www.nature.com/articles/s41586-025-09429-6",
    "title": "Journal article page",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/martinpacesa/BindCraft/releases",
    "title": "Official repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://news.epfl.ch/news/an-open-source-ai-platform-to-democratize-protei-2/",
    "title": "University news release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://adaptyvbio.substack.com/p/protein-design-competition-results",
    "title": "Competition results blog (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.event_kind",
    "was": "2026-04-17 was recorded as the journal publication",
    "now": "That date is the v1.5.3 small fix, so the kind becomes patch; the journal version online 2025-08-27 is listed separately",
    "source_url": "https://github.com/martinpacesa/BindCraft/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "The 46.3% average was labelled a second-hand summary",
    "now": "The figure is stated directly in the journal body text, so the second-hand label is dropped; the rate is the share of tested designs that bound",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.code_release",
   "benchmarks_reported[0]",
   "open_status.weights"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-bindcraft/",
   "en": "/bio/benchmark/models/bindcraft/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-bioemu-2",
  "kind": "model",
  "short_slug": "bioemu-2",
  "names": {
   "display": "BioEmu-2",
   "full": null,
   "aliases": [
    "BioEmu-2",
    "bioemu2",
    "bioemu2-preview"
   ],
   "ambiguous_aliases": [
    "BioEmu"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "BioEmu-1",
     "id_or_null": null,
     "first_release": {
      "date": "2025",
      "precision": "year",
      "year_hint": null,
      "basis": "Year of journal publication",
      "event_kind": "journal-publication",
      "timestamp_utc": null,
      "source_url": "https://doi.org/10.1126/science.adv9817",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "BioEmu-2",
     "id_or_null": "model-bioemu-2",
     "first_release": {
      "date": "2026-10",
      "precision": "month",
      "year_hint": null,
      "basis": "Release month stated on the model card",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://huggingface.co/microsoft/bioemu2",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "This record covers the second generation only; the first handles single proteins and its results may not be carried over"
  },
  "family_null_reason": null,
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "All-atom diffusion model sampling structural ensembles of complexes and nucleic acids",
  "org": {
   "names": [
    "Microsoft Research AI for Science"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US",
   "DE"
  ],
  "model_class": "Structural ensembles (all-atom diffusion)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Weights and code under MIT, with the developer stating research use only and advising against commercial or real-world use without further testing"
  },
  "first_release": {
   "date": "2026-10",
   "precision": "month",
   "year_hint": null,
   "basis": "Release month on the model card, with training in August and September that year",
   "event_kind": "new-weights",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/microsoft/bioemu2",
   "verification": "partial",
   "note": "No primary source gives the exact day"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-10",
    "precision": "month",
    "year_hint": null,
    "basis": "Release month on the model card",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/microsoft/bioemu2",
    "verification": "partial",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2026-10-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Pre-release tag v0.1.0rc0",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/microsoft/bioemu2",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-10",
    "precision": "month",
    "year_hint": null,
    "basis": "Weights published with the public preview",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/microsoft/bioemu2",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-10-20",
   "precision": "day",
   "year_hint": null,
   "basis": "Pre-release tag published",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/microsoft/bioemu2",
   "verification": "verified",
   "description": "Pre-release tag v0.1.0rc0 on 20 October 2026, with the model hub last modified the day before; repository activity through 16 November 2026 is automated dependency bumps.",
   "note": "The site previously recorded no releases or tags, which the releases page contradicts"
  },
  "status": "active",
  "status_note": "Public preview and a pre-release tag within 12 months",
  "what_it_does": "Samples equilibrium structural ensembles of biomolecules and can predict folded structures, extending the first generation from single proteins to multi-chain complexes, ligands and nucleic acids, taking structure files, sequences or multiple sequence alignments as input. The model card states it does not generate sequences or predict affinity, that sampling frequencies are not calibrated probabilities and that outputs are hypotheses rather than measurements. It shares the ensemble route with OpenComplex2 and differs from co-folding models such as AlphaFold3 in returning a distribution rather than one structure.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 189,700 systems from a public structure archive, plus monomer, homodimer and heterodimer entries from a public predicted-structure archive and one multimer set",
    "source_url": "https://huggingface.co/microsoft/bioemu2"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The model card states only public data and the developer's own data are used",
    "source_url": "https://huggingface.co/microsoft/bioemu2"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "The in-house data are molecular dynamics trajectories, computationally generated rather than wet-lab",
    "source_url": "https://huggingface.co/microsoft/bioemu2"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Evaluation areas listed on the model card",
    "category": "own-task-set",
    "detail": "Lists folding, multi-conformation sampling, a dynamics-emulation holdout and large systems, with metric names but no reported values",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a foundation model for structures and ensembles of complexes, ligands and nucleic acids; the model card gives no quantitative results and states the limitations are not quantitatively characterised.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "It has no result among the 44 benchmarks here and no third-party evaluation was found",
  "background": {
   "org_profile": "The developer is a large technology company's science-AI research division, formed in 2022 from existing teams across Europe, North America and Asia with two new European labs, funded from the company research budget.",
   "funding": "No separate funding; it is internal corporate research.",
   "products_partnerships": "The first generation was published in a general science journal in 2025, released under MIT and offered through the company's model platform, with a July 2026 customer story describing a pharmaceutical company using it there; whether the second generation is on the platform is not officially stated.",
   "related_models": [
    "model-opencomplex2",
    "model-alphafold3",
    "model-boltz-2"
   ]
  },
  "reading_notes": [
   "For now there is a model but no numbers: with no reported values it cannot be called better or worse than its predecessor or its peers.",
   "The new complex and nucleic-acid capability has no public evaluation, and the pose benchmarks here score single structures, which does not match its goal.",
   "The dynamics trajectories are produced in-house, so their coverage is the ceiling, and the model card itself says the training set leans towards well-studied protein families.",
   "Sampling frequencies are not calibrated probabilities, so they cannot be read as free energies and downstream use needs its own validation."
  ],
  "caveats": [
   "Neither model card nor repository reports any evaluation value.",
   "First release is month-level, with no primary day-level source.",
   "Weights are open but the developer limits them to research use.",
   "The in-house training data are computed trajectories, not wet-lab data.",
   "Results from the first generation may not be carried over to this one."
  ],
  "sources": [
   {
    "url": "https://huggingface.co/microsoft/bioemu2",
    "title": "Model card",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/microsoft/bioemu2",
    "title": "Code repository and releases page",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1126/science.adv9817",
    "title": "First-generation journal paper 2025",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-11-16 (automated dependency-upgrade commit)",
    "now": "2026-10-20 (pre-release tag v0.1.0rc0)",
    "source_url": "https://github.com/microsoft/bioemu2",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.first_public_evidence",
   "release_milestones.weights_release"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-bioemu-2/",
   "en": "/bio/benchmark/models/bioemu-2/"
  },
  "names_full_null_reason": "A product name plus generation, with no expanded form in the repository",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-biomni",
  "kind": "model",
  "short_slug": "biomni",
  "names": {
   "display": "Biomni",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "Biomni",
    "Biomni Lab",
    "Biomni-R0"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "General biomedical agent: an LLM plus a tool library that plans and runs bioinformatics analyses",
  "org": {
   "names": [
    "Stanford University",
    "Phylo"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": "Built in an academic lab; commercialized by the startup Phylo",
  "country": [
   "US"
  ],
  "model_class": "Biomedical agent",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Open-source platform; Biomni Lab is a commercial research preview"
  },
  "first_release": {
   "date": "2025-06-02",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv preprint v1 (Crossref issued 2025-06-02)",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2025.05.30.656746",
   "verification": "verified",
   "note": "The open-source repository may predate the preprint; unverified"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-06-02",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1 (Crossref issued date)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.05.30.656746",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-08-20",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Science publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1126/science.adz4351",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-02-03",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Biomni Lab launch, announced with the Phylo seed round",
    "timestamp_utc": null,
    "source_url": "https://pulse2.com/phylo-13-5-million-seed-funding-raised-and-biomni-lab-launched",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-08-20",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Science journal publication",
   "description": "Published in Science; Biomni Lab launched 2026-02-03 and GitHub v0.0.8 came out 2025-10-27",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1126/science.adz4351",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication and product launch within 12 months",
  "what_it_does": "Combines a large language model with tools, software packages and databases to plan and run bioinformatics analyses and experiment design. Besides the open-source platform, Phylo offers the commercial Biomni Lab; the companion Biomni-R0 is trained on agent trajectories.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Tools, packages and databases are public resources",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "One cloning experiment used for validation",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.05.30.656746v1.full"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Cloning task (single)",
    "modality": "molecular biology protocol",
    "targets": null,
    "designs_tested": 1,
    "hits": 1,
    "denominator_basis": "A single cloning task; success = colonies on selective media plus sequence-verified insertion",
    "success_rate": null,
    "affinity_best": null,
    "version": "bioRxiv v1",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.05.30.656746v1.full",
    "date": null,
    "verification": "verified",
    "note": "System-generated protocols used unmodified; colony counts not given; developer-run, not an independent replication"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-18-labbench",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "LAB-Bench DbQA/SeqQA subsets",
    "source_url": "https://doi.org/10.1101/2025.05.30.656746",
    "confidence": "low",
    "verification": "partial",
    "note": "Developer-reported on DbQA/SeqQA subsets only; protocol differs from official"
   },
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "BixBench full + verified-50",
    "source_url": "https://phylo.bio/blog/evaluating-ai-agents-in-biology",
    "confidence": "medium",
    "verification": "partial",
    "note": "Run by the developer Phylo (full set and verified-50 subset); Phylo competes with the benchmark maker"
   },
   {
    "benchmark_id": "bench-41-bioml-bench",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.09.01.673319",
    "confidence": "medium",
    "verification": "partial",
    "note": "Benchmark authors re-ran Biomni; site matrix says self, should be third-party (open item)"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": "Developers report near-expert performance on several biomedical tasks and zero-shot handling of new tasks",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "In a BioML-bench re-run by its authors, Biomni was one of few agents to complete all tasks",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Started at the Stanford AI Lab (2024-06); the commercial entity Phylo is based in South San Francisco",
   "funding": "Phylo announced a $13.5M seed round on 2026-02-03, led by Andreessen Horowitz and Menlo Ventures' Anthology Fund, with Anthropic",
   "funding_null_reason": null,
   "products_partnerships": "Commercial product Biomni Lab (research preview)",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-kosmos",
    "model-robin",
    "model-txagent"
   ]
  },
  "reading_notes": [
   "Its LAB-Bench and BixBench numbers are developer- or company-run on modified protocols",
   "The only wet-lab item is one cloning task: a demonstration, not systematic validation"
  ],
  "caveats": [
   "BixBench results were published by a company that competes with the benchmark maker",
   "The BioML-bench pairing source type on the site matrix is still to be changed to third-party"
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1126/science.adz4351",
    "title": "Journal article DOI 10.1126/science.adz4351",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2025.05.30.656746",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1126/science.adz4351",
    "title": "Crossref record 10.1126/science.adz4351",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/snap-stanford/Biomni/releases",
    "title": "GitHub repository snap-stanford/Biomni",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://pulse2.com/phylo-13-5-million-seed-funding-raised-and-biomni-lab-launched",
    "title": "Official page pulse2.com",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://phylo.bio/blog/evaluating-ai-agents-in-biology",
    "title": "Phylo blog: evaluating AI agents in biology",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.09.01.673319",
    "title": "BioML-bench preprint",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-biomni/",
   "en": "/bio/benchmark/models/biomni/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-bioreason",
  "kind": "model",
  "short_slug": "bioreason",
  "names": {
   "display": "BioReason",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "BioReason"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "genomics-rna",
  "domains_secondary": [
   "text-agents"
  ],
  "one_liner": "Multimodal model feeding a DNA foundation model into an LLM for pathway reasoning and variant effects",
  "org": {
   "names": [
    "University of Toronto",
    "Vector Institute",
    "University Health Network"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CA"
  ],
  "model_class": "DNA-language multimodal reasoning",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code on GitHub, datasets on Hugging Face"
  },
  "first_release": {
   "date": "2025-05-29",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2505.23579 v1 submitted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2505.23579",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-05-29",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2505.23579",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-07-19",
   "precision": "day",
   "event_kind": "successor-model",
   "basis": "bioRxiv v2 of the successor BioReason-Pro",
   "description": "Successor BioReason-Pro: bioRxiv v1 2026-03-20, v2 2026-07-19; BioReason itself has arXiv v2 and NeurIPS 2025 proceedings",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.03.19.712954",
   "verification": "verified",
   "note": null
  },
  "status": "superseded",
  "status_note": "Successor BioReason-Pro exists",
  "what_it_does": "Feeds representations from a DNA foundation model (Evo 2 or Nucleotide Transformer) into a Qwen3 language model for KEGG pathway reasoning and variant-effect prediction, with reasoning given as text.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "KEGG pathways plus ClinVar, OMIM, dbSNP and COSMIC annotations; gnomAD benign controls",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-11-clinvar",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": null,
    "source_url": "https://arxiv.org/abs/2505.23579v2",
    "confidence": "high",
    "verification": "verified",
    "note": "Developer-reported on a self-built KEGG variant task, not the ClinVar hold-out protocol"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "Academic project of the University of Toronto, Vector Institute and University Health Network; accepted at NeurIPS 2025",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": null,
   "products_partnerships_null_reason": "No product or partnership found",
   "related_models": [
    "model-evo-2",
    "model-scgpt-geneformer"
   ]
  },
  "reading_notes": [
   "Its only pairing is a self-built task, not comparable with the official ClinVar protocol",
   "2026 activity comes from the successor BioReason-Pro, not a new BioReason version"
  ],
  "caveats": [
   "The BioReason arXiv v2 date (2025-10-24) was not re-verified",
   "No wet-lab validation"
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2505.23579",
    "title": "arXiv preprint 2505.23579",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/bowang-lab/BioReason-Pro",
    "title": "GitHub repository bowang-lab/BioReason-Pro",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/bowang-lab/BioReason",
    "title": "GitHub repository bowang-lab/BioReason",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.52202/085713-3728",
    "title": "Journal article DOI 10.52202/085713-3728",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-08-20 (successor repository commit)",
    "now": "2026-07-19 (BioReason-Pro bioRxiv v2)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.03.19.712954",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "Successor repository creation and commits described as the update",
    "now": "Changed to the successor preprint events; repository commits are not releases",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.03.19.712954",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-bioreason/",
   "en": "/bio/benchmark/models/bioreason/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-boltz-1-1x",
  "kind": "model",
  "short_slug": "boltz-1-1x",
  "names": {
   "display": "Boltz-1 / Boltz-1x",
   "full": null,
   "full_null_reason": "Both members use short names only; no expanded name",
   "aliases": [
    "Boltz-1",
    "Boltz-1x"
   ],
   "ambiguous_aliases": [
    "Boltz"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Boltz-1",
     "id_or_null": null,
     "first_release": {
      "date": "2024-11-17",
      "precision": "day",
      "event_kind": "code-release",
      "basis": "GitHub release v0.1.0",
      "timestamp_utc": null,
      "source_url": "https://github.com/jwohlwend/boltz/releases",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "Boltz-1x",
     "id_or_null": null,
     "first_release": {
      "date": "2025-04-26",
      "precision": "day",
      "event_kind": "new-version",
      "basis": "GitHub release v1.0.0, adding inference-time physical steering",
      "timestamp_utc": null,
      "source_url": "https://github.com/jwohlwend/boltz/releases",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "This record covers Boltz-1 and Boltz-1x only; Boltz-2 and BoltzGen have their own records and are not merged in."
  },
  "family_null_reason": null,
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "First fully open co-folding model approaching AF3, with 1x adding physical-validity steering",
  "org": {
   "names": [
    "MIT",
    "Boltz PBC"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": "Started in a university lab, now maintained by a public benefit corporation",
  "country": [
   "US"
  ],
  "model_class": "Co-folding (protein / nucleic acid / small molecule)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Code and weights are both MIT-licensed with commercial use allowed; the technical report says the dataset and evaluation sets are released under MIT too"
  },
  "first_release": {
   "date": "2024-11-17",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "Boltz-1's first GitHub release v0.1.0; preprint 2024-11-20",
   "timestamp_utc": null,
   "source_url": "https://github.com/jwohlwend/boltz/releases",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-11-20",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2024.11.19.624167",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-11-17",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub release v0.1.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/jwohlwend/boltz/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2024-11-17",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights published with the first release under MIT",
    "timestamp_utc": null,
    "source_url": "https://github.com/jwohlwend/boltz",
    "verification": "partial",
    "note": "Based on the repository README; no separate release page"
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-06-06",
   "precision": "day",
   "event_kind": "successor-model",
   "basis": "Successor in the same product line released",
   "description": "Boltz-2 shipped as the same-line successor (adding affinity prediction); Boltz-1x's steering was v1.0.0 on 2025-04-26",
   "timestamp_utc": null,
   "source_url": "https://github.com/jwohlwend/boltz/releases",
   "verification": "verified",
   "note": null
  },
  "status": "superseded",
  "status_note": "Superseded by its successor but still a common open control",
  "what_it_does": "The first fully open co-folding model to approach AlphaFold 3 on public evaluations, covering protein, nucleic-acid and small-molecule complexes; 1x adds steering potentials at inference to cut stereochemical and steric clashes. It is a standard open control in papers such as Chai-1 and Protenix.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB plus distillation; no per-database sizes, dataset released under MIT",
    "source_url": "https://github.com/jwohlwend/boltz"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Neither member reports wet-lab work",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "PoseBusters V2, best@5",
    "source_url": "https://arxiv.org/abs/2510.24670",
    "confidence": "medium",
    "verification": "partial",
    "note": "The figures come from a third-party re-run; the developers do not self-report here"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "Runs N' Poses, best@5",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): figures quoted from a third-party paper"
   },
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer leaderboard",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "high",
    "verification": "verified",
    "note": "The maintainer leaderboard lists Boltz-1"
   },
   {
    "benchmark_id": "bench-33-posex",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "benchmark authors' own sets",
    "source_url": "https://arxiv.org/html/2505.01700v3",
    "confidence": "medium",
    "verification": "partial",
    "note": "Evaluated by the benchmark's authors on their own sets"
   }
  ],
  "other_evaluations": [
   {
    "label": "Inclusion in third-party evaluation suites",
    "category": "third-party-evaluation",
    "detail": "Two third-party evaluation efforts (a 2026 journal paper and a conference paper) include it in multi-method comparisons, relayed second-hand",
    "source_type": "third"
   },
   {
    "label": "Qualitative statements in the technical report",
    "category": "comparison-vs-baseline-only",
    "detail": "The abstract and README give only qualitative statements of parity with contemporary closed models, with no evaluation values",
    "source_type": "self"
   },
   {
    "label": "Qualitative finding from a third-party reproduction",
    "category": "third-party-evaluation",
    "detail": "A reproduction report says it works for most orthosteric ligands, without summary metrics",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: the first fully open model to match contemporary closed co-folding quality on public evaluations, with code, weights and dataset under a commercially usable licence.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Every figure on its page comes from third-party re-runs, plus inclusion in a maintainer leaderboard; the developers report no scores themselves, so reproducibility is the firmest evidence.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The model comes from a US university lab and is now maintained by a public benefit corporation spun out of it; the company launched publicly on 2026-01-08, says it does not develop its own drugs and charges for software and platform access, with headcount undisclosed.",
   "funding": "2026-01 a $28M seed round co-led by three firms with two further funds and angels participating; valuation not officially disclosed.",
   "funding_null_reason": null,
   "products_partnerships": "Two lines, open models plus closed enterprise models; a 2026-01 multi-year pharma collaboration refines custom models on that company's historical data while the base models stay open; on 2026-06-16 two further proprietary models and a hosted API were announced (secondary coverage only, unverified).",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-boltz-2",
    "model-boltzgen",
    "model-alphafold3",
    "model-chai-1",
    "model-protenix",
    "model-openfold3"
   ]
  },
  "reading_notes": [
   "The gap between the two members comes mainly from physical-validity steering, so the validity criterion matters a lot and comparisons must name which member is meant.",
   "All concrete figures are third-party re-runs under differing protocols and cannot be read as a strict ranking.",
   "Its value lies in reproducibility and fine-tunability rather than leaderboard position.",
   "The successor's release is a same-line event and does not mean these two members were themselves updated."
  ],
  "caveats": [
   "Family boundary: Boltz-1 and Boltz-1x only; the successor and the design model stand alone.",
   "The developers self-report no benchmark values; all figures are third-party.",
   "One pairing rests only on the site's own page and displays as unverified.",
   "The weights-release date rests on the README with no separate release page (partially verified).",
   "The two proprietary models of 2026-06-16 have secondary coverage only and are not adopted as facts here."
  ],
  "sources": [
   {
    "url": "https://github.com/jwohlwend/boltz",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2024.11.19.624167",
    "title": "Boltz-1 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/jwohlwend/boltz/releases",
    "title": "Official repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2510.24670",
    "title": "Third-party re-run paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2505.01700v3",
    "title": "Re-run paper by the benchmark's authors",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://www.pulse.bot/biotech/news/boltz-pbc-launches-with-28m-to-democratize-ai-platforms-for-drug-discovery-aecd43c8-8e27-4e04-b86a-4d32e199f425",
    "title": "Coverage of the maintaining public benefit corporation's launch (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "family",
    "was": "The two members were written up as a single 2024-11 release",
    "now": "The family boundary is stated: Boltz-1 on 2024-11-17 and the steering version Boltz-1x on 2025-04-26",
    "source_url": "https://github.com/jwohlwend/boltz/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "First release taken from the preprint date",
    "now": "Changed to the earlier first code release, with the preprint kept as a separate milestone",
    "source_url": "https://github.com/jwohlwend/boltz/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Third-party re-run figures were displayed as developer self-reports",
    "now": "Two rows become comparison-only citations and are not counted as established relations",
    "source_url": null,
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[1]",
   "release_milestones.weights_release",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-boltz-1-1x/",
   "en": "/bio/benchmark/models/boltz-1-1x/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-boltz-2",
  "kind": "model",
  "short_slug": "boltz-2",
  "names": {
   "display": "Boltz-2",
   "full": null,
   "full_null_reason": "Official materials use Boltz-2 only; no expanded name",
   "aliases": [
    "Boltz-2"
   ],
   "ambiguous_aliases": [
    "Boltz"
   ]
  },
  "family": null,
  "family_null_reason": "Covers Boltz-2 only; Boltz-1/1x and BoltzGen have their own records",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Open model that adds a small-molecule binding-affinity head on top of co-folding",
  "org": {
   "names": [
    "MIT",
    "Recursion",
    "Boltz PBC"
   ],
   "global_entity_ids": [
    "recursion-pharmaceuticals"
   ]
  },
  "org_type": "mixed",
  "org_type_note": "Jointly released by a university lab and a listed biotech company, now maintained by a public benefit corporation",
  "country": [
   "US"
  ],
  "model_class": "Co-folding plus affinity prediction",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Weights and inference code are MIT-licensed; as of 2026-11-16 the repository still marks the training code and evaluation scripts as coming soon"
  },
  "first_release": {
   "date": "2025-06-06",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release v2.0.0; preprint 2025-06-18",
   "timestamp_utc": null,
   "source_url": "https://github.com/jwohlwend/boltz/releases",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-06-18",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.06.14.659707",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-06-06",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub release v2.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/jwohlwend/boltz/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-06-06",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights published with v2.0.0 under MIT",
    "timestamp_utc": null,
    "source_url": "https://github.com/jwohlwend/boltz",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-06-06",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Joint release note by the institute and the partner company",
    "timestamp_utc": null,
    "source_url": "https://www.csail.mit.edu/news/mit-releases-breakthrough-protein-binding-affinity-model-expanding-role-ai-drug-discovery",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-06-16",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Product-line event: two new models and a hosted API went live (secondary sources only)",
   "description": "The developers announced two new models and a hosted API running this model; the open repository's last release was 2025-09-08 and the weights themselves are unchanged (unverified)",
   "timestamp_utc": null,
   "source_url": "https://www.rdworldonline.com/boltz-built-its-drug-discovery-api-for-agents-as-much-as-for-people/",
   "verification": "unverified",
   "note": "The date appears only in secondary coverage; the developers' own page was not retrieved (unverified)"
  },
  "status": "maintained",
  "status_note": "No open-repository release since 2025-09; the 2026 product event is unverified",
  "what_it_does": "On the structure side it adds method conditioning and more control over the predecessor; on the affinity side a head on top of the structure module outputs binder or non-binder plus binding strength for hit finding and hit-to-lead ranking. That affinity side is what separates it from Boltz-1/1x and Chai-1, and Nesso-1, TerraBind and AQAffinity reuse its public-data recipe.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB (to 2023-06-01), MD ensembles, distillation plus public affinity libraries such as ChEMBL; totals differ by source",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "None in the public version; from 2026-01 a pharma partner's historical data is used for refinement, with no statement on the public version",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No new wet lab of its own; a partner's eight internal assays serve as a held-out blind test and one prospective item is a simulation",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Partner blinded assays",
    "modality": "small molecule affinity",
    "targets": 8,
    "designs_tested": null,
    "hits": 3,
    "denominator_basis": "Eight internal partner assays, using existing experimental data as a held-out blind test rather than wet lab run for this model",
    "success_rate": "3/8 assays above the stated correlation threshold",
    "affinity_best": null,
    "version": "journal paper",
    "source_type": "partner-lab",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/",
    "date": {
     "date": "2025-06-18",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "Published with the paper",
     "timestamp_utc": null,
     "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "note": "Partner-lab executed and developer-compiled, so not an independent replication; the correlation threshold is a metric and is not shown here"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "affinity track, retrospective (2 targets / 140 pairs)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/",
    "confidence": "high",
    "verification": "verified",
    "note": "Retrospective use of the affinity data after the fact, not blind participation"
   },
   {
    "benchmark_id": "bench-12-dude",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": "decoy-set screening in a third-party paper",
    "source_url": "https://pubs.rsc.org/en/content/articlelanding/2026/sc/d5sc06481c",
    "confidence": "low",
    "verification": "unverified",
    "note": "The third-party paper uses a different decoy set, membership of this family is unclear, and the account is second-hand (unverified)"
   },
   {
    "benchmark_id": "bench-27-polaris",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": null,
    "source_url": null,
    "confidence": "low",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): the page claims parity with the top entries without fine-tuning, gives no values, and neither participation nor protocol is primary-verified"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "Runs N' Poses",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): quoted from a third-party paper and flagged as not directly comparable"
   },
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer run on post-2024-01 subset",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "medium",
    "verification": "partial",
    "note": "Run by the maintainers on a later subset and treated as reference only because of the later training cutoff"
   },
   {
    "benchmark_id": "bench-34-abbibench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "registered interface-score entry",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): the entry is an interface score, not structure accuracy"
   }
  ],
  "other_evaluations": [
   {
    "label": "Free-energy calculation comparison subset",
    "category": "own-task-set",
    "detail": "Correlation against a physics-based method on a four-target subset, self-reported; scores are not shown here",
    "source_type": "self"
   },
   {
    "label": "Hit-to-lead and screening sets, self-reported",
    "category": "own-task-set",
    "detail": "Self-reported results on a public hit-to-lead dataset and a high-throughput screening dataset, neither a site benchmark",
    "source_type": "self"
   },
   {
    "label": "Third-party prospective evaluation",
    "category": "third-party-prospective-evaluation",
    "detail": "A prospective study on about 557 co-crystal complexes finds its affinity score best for early enrichment while its poses trail the contemporary closed model",
    "source_type": "third"
   },
   {
    "label": "Third-party protein-protein affinity study",
    "category": "third-party-evaluation",
    "detail": "A paper finds it still below a sequence baseline after fine-tuning",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: near physics-based free-energy accuracy on a four-target subset at about three orders of magnitude lower cost, matching the top entries of a public challenge without fine-tuning.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Third-party prospective work finds the affinity score useful for enrichment but insufficient alone for virtual screening; other work argues screening results are inflated by training overlap and that it trails a sequence baseline on protein-protein tasks.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "At release it was co-authored by two research units of a US university and a listed biotech company that supplied data and compute; it is now maintained by a public benefit corporation spun out of that university, which says it does not develop its own drugs.",
   "funding": "The maintainer launched publicly on 2026-01-08 with a $28M seed round co-led by three firms; valuation undisclosed.",
   "funding_null_reason": null,
   "products_partnerships": "A platform product launched the same day, alongside a pharma collaboration building custom structure, affinity and biologics-design models on that company's data (custom development, per second-hand analysis); the two new models and hosted API of 2026-06-16 appear only in secondary coverage (unverified).",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-boltz-1-1x",
    "model-boltzgen",
    "model-nesso-1",
    "model-terrabind",
    "model-aqaffinity",
    "model-isodde"
   ]
  },
  "reading_notes": [
   "The free-energy comparison holds only on a four-target subset, and only high-sequence-identity targets were removed; overlap with other evaluation sets is not reported set by set.",
   "Only a minority of the eight partner blind assays clear the threshold, so performance on new targets is unstable and the subset result cannot be extrapolated.",
   "Because of its later cutoff the structure side is often excluded from PoseBusters-style comparisons, so no claim either way about structure quality follows.",
   "The 2026 product event comes from secondary coverage, and the weights themselves are unchanged since 2025-09."
  ],
  "caveats": [
   "Training code and evaluation scripts are still marked coming soon in the repository, so full reproduction of the paper is not yet possible.",
   "The most recent update event has secondary sources only and must display as unverified.",
   "Three pairings rest only on the site's own pages (unverified).",
   "One pairing uses a different decoy set and its membership of that family is still undecided.",
   "The wet-lab item is a held-out blind test on a partner's existing data, neither new experiments nor an independent replication."
  ],
  "sources": [
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.06.14.659707",
    "title": "The model's preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/",
    "title": "Paper full text (public archive)",
    "kind": "paper",
    "date": null
   },
   {
    "url": "https://github.com/jwohlwend/boltz",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/jwohlwend/boltz/releases",
    "title": "Official repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.csail.mit.edu/news/mit-releases-breakthrough-protein-binding-affinity-model-expanding-role-ai-drug-discovery",
    "title": "Institute release note",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://elifesciences.org/reviewed-preprints/110475.pdf",
    "title": "Third-party prospective evaluation preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2510.24670",
    "title": "Third-party re-run paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://www.rdworldonline.com/boltz-built-its-drug-discovery-api-for-agents-as-much-as-for-people/",
    "title": "Coverage of the 2026 product event (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "The 2026-06-16 product event was presented as settled fact",
    "now": "Marked unverified: secondary coverage only, with the developers' own page not retrieved",
    "source_url": "https://www.rdworldonline.com/boltz-built-its-drug-discovery-api-for-agents-as-much-as-for-people/",
    "date_found": "2026-11-16"
   },
   {
    "field": "other_evaluations",
    "was": "A free-energy correlation value relayed by a second-hand page",
    "now": "Corrected to the paper's own figure, and this record shows no concrete scores",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "The partner assays were treated as this model's own wet-lab validation",
    "now": "Reclassified as a held-out blind test on existing data, partner-lab executed, with one prospective item a simulation rather than an experiment",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12262699/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "training_data.b_private_licensed",
   "release_milestones.weights_release",
   "benchmarks_reported[1]",
   "benchmarks_reported[2]",
   "benchmarks_reported[3]",
   "benchmarks_reported[5]",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-boltz-2/",
   "en": "/bio/benchmark/models/boltz-2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-boltzgen",
  "kind": "model",
  "short_slug": "boltzgen",
  "names": {
   "display": "BoltzGen",
   "full": null,
   "aliases": [
    "BoltzGen"
   ],
   "ambiguous_aliases": [
    "Boltz"
   ]
  },
  "family": null,
  "family_null_reason": "Record covers BoltzGen only; the closed successor BoltzProt-1 is unverified and not merged in",
  "domain": "protein",
  "domains_secondary": [
   "generation",
   "structure"
  ],
  "one_liner": "Open general-purpose binder design model built on a co-folding architecture",
  "org": {
   "names": [
    "MIT Jameel Clinic",
    "Boltz PBC"
   ],
   "global_entity_ids": []
  },
  "org_type": "open-source-community",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (co-folding generation: protein / peptide / nanobody)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code and weights both public on GitHub; licence text not retrieved this round, so both terms are left empty"
  },
  "first_release": {
   "date": "2025-10-26",
   "precision": "day",
   "year_hint": null,
   "basis": "First public code and weights (first repository commit, corroborated by two technology outlets)",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/HannesStark/boltzgen",
   "verification": "partial",
   "note": "The site said 2025-11; the manuscript sits on the release page and repository, the bioRxiv entry (2025-11-24) is later"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-10-26",
    "precision": "day",
    "year_hint": null,
    "basis": "First repository commit, open-source release",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/HannesStark/boltzgen",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-11-24",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv v1 (10.1101/2025.11.20.689494), not journal-published",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12697729/",
    "verification": "partial",
    "note": "Corrects the site's claim that no bioRxiv entry exists"
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-10-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Code public on GitHub",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/HannesStark/boltzgen",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2025-10-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights released with the code",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://github.com/HannesStark/boltzgen",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-10-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Research-institute release page",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://jclinic.mit.edu/boltzgen/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-05-01",
   "precision": "day",
   "year_hint": null,
   "basis": "GitHub release v0.3.2",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://github.com/HannesStark/boltzgen/releases",
   "verification": "verified",
   "description": "v0.3.2 fixes binding_types; v0.3.1 (2026-03-06) added per-residue constraints and ipSAE, v0.3.0 (2026-02-10) symmetric inverse folding and covalent bonds; commits continue to 2026-11-01",
   "note": null
  },
  "status": "active",
  "status_note": "Version releases and ongoing commits within 12 months",
  "what_it_does": "Puts structure prediction and design in one generative framework, generating binders conditioned on protein, peptide or small-molecule targets under design constraints. Shares lineage with Boltz-2 but is not a structure predictor; unlike the backpropagation route of BindCraft it is a trained generative model, and unlike the closed Latent-X/X2 and Chai-2 its weights are open and reproducible.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB, AFDB and similar public structure data (Boltz family)",
    "source_url": "https://github.com/HannesStark/boltzgen"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "None for this model; a later pharma collaboration uses that company's historical data, outside this model",
    "source_url": "https://jclinic.mit.edu/boltzgen/"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Designs assayed in the wet lab, run by a CRO with orthogonal confirmation by a second vendor",
    "source_url": "withheld: author-hosted manuscript"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Novel hard targets, nanobody arm",
    "targets": 10,
    "designs_tested": 15,
    "hits": 6,
    "denominator_basis": "Ten novel targets with no bound structure in the PDB (1%-27.9% identity to any protein seen bound); 60,000 nanobodies designed per target, top 15 per modality assayed",
    "success_rate": "6/10",
    "affinity_best": null,
    "version": "manuscript (paper v2 lineage)",
    "source_type": "self",
    "source_url": "withheld: author-hosted manuscript",
    "date": {
     "date": "2025-11-24",
     "precision": "day",
     "year_hint": null,
     "basis": "Manuscript version (same period as bioRxiv v1)",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "withheld: author-hosted manuscript",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "CRO-executed with orthogonal confirmation by a second vendor; not an independent evaluation",
    "unverified_marker": null,
    "note": "Denominator follows the manuscript (10), not the release page (9)"
   },
   {
    "campaign": "Novel hard targets, miniprotein arm",
    "targets": 10,
    "designs_tested": 15,
    "hits": 5,
    "denominator_basis": "Same ten targets; 60,000 miniproteins designed per target, top 15 assayed",
    "success_rate": "5/10",
    "affinity_best": null,
    "version": "manuscript (paper v2 lineage)",
    "source_type": "self",
    "source_url": "withheld: author-hosted manuscript",
    "date": {
     "date": "2025-11-24",
     "precision": "day",
     "year_hint": null,
     "basis": "Manuscript version",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "withheld: author-hosted manuscript",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "CRO-executed with orthogonal confirmation by a second vendor",
    "unverified_marker": null,
    "note": "A second 6/9 figure once published had no source and is removed"
   },
   {
    "campaign": "Developer release page figure (nanobody)",
    "targets": 9,
    "designs_tested": null,
    "hits": 6,
    "denominator_basis": "Release page states nine novel targets",
    "success_rate": "6/9",
    "affinity_best": null,
    "version": "developer release page",
    "source_type": "self",
    "source_url": "https://jclinic.mit.edu/boltzgen/",
    "date": {
     "date": "2025-10-26",
     "precision": "day",
     "year_hint": null,
     "basis": "Release page published",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://jclinic.mit.edu/boltzgen/",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Developer's own account",
    "unverified_marker": null,
    "note": "Kept only as that version's own figure; it disagrees with the manuscript denominator (10) and must never be merged with it"
   },
   {
    "campaign": "Known easy targets (five-target set from another model)",
    "targets": 5,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Five easy targets, each present at 100% identity in at least one bound PDB structure",
    "success_rate": "nanobody 4/5 (1:1 binders confirmed for 3); miniprotein 3/5",
    "affinity_best": null,
    "version": "manuscript",
    "source_type": "self",
    "source_url": "withheld: author-hosted manuscript",
    "date": {
     "date": "2025-11-24",
     "precision": "day",
     "year_hint": null,
     "basis": "Manuscript version",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "withheld: author-hosted manuscript",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "CRO-executed",
    "unverified_marker": null,
    "note": "The site's 4/5 holds for nanobodies only"
   },
   {
    "campaign": "Partner design campaigns (small molecule / peptide / enzyme / disordered region)",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "The abstract cites eight design campaigns with functional and affinity readouts across several targets; the manuscript does not describe all of them as external partner-lab work",
    "success_rate": null,
    "affinity_best": null,
    "version": "manuscript",
    "source_type": "self",
    "source_url": "withheld: author-hosted manuscript",
    "date": {
     "date": "2025-11-24",
     "precision": "day",
     "year_hint": null,
     "basis": "Manuscript version",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "withheld: author-hosted manuscript",
     "verification": "unverified",
     "note": null
    },
    "verification": "unverified",
    "executor": "Partner laboratories, attribution unverified",
    "unverified_marker": "unverified",
    "note": "No figures given: the published 'external labs' framing is unsupported"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Own novel-target wet-lab evaluation",
    "category": "wet-lab-validation",
    "detail": "Ten novel targets, two modalities, top 15 designs per modality assayed as the developers' own evaluation set",
    "source_type": "self"
   },
   {
    "label": "Known-target check",
    "category": "wet-lab-validation",
    "detail": "Reuses another model's five published easy targets as a control",
    "source_type": "self"
   },
   {
    "label": "Partner laboratory campaigns",
    "category": "case-study",
    "detail": "Functional and affinity readouts across campaigns; scale and executor attribution unverified",
    "source_type": "self"
   },
   {
    "label": "Parity with folding models",
    "category": "comparison-vs-baseline-only",
    "detail": "The release page claims parity with contemporary folding models and gives no metrics",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one open model designs protein, nanobody and peptide binders across target types and reaches nanomolar binders on novel targets without known binders.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Wet-lab work was CRO- and partner-executed but compiled by the developers; no independent replication or third-party evaluation found",
  "background": {
   "org_profile": "The model comes from a US university AI-health clinic together with the Boltz team. Boltz PBC, a public benefit corporation, launched 2026-01-08 in Cambridge, Massachusetts; it pairs open research models with closed enterprise models and does not develop therapeutics itself. No public headcount.",
   "funding": "$28M seed led by Amplify, a16z and Zetta Venture Partners with further funds and angels; a $125M valuation appears only in second-hand reports and is not adopted.",
   "products_partnerships": "2026-01-08 multi-year collaboration with a large pharma company (fine-tuning on its historical data, base models stay open, terms undisclosed); 2026-06-18 collaboration with another pharma company granting model and API access, terms undisclosed. The closed successor BoltzProt-1 (2026-06-16) is said to raise hit rates nearly threefold, while a second-hand report says 2.4x; the figures conflict and are unverified.",
   "related_models": [
    "model-boltz-1-1x",
    "model-boltz-2",
    "model-bindcraft",
    "model-latent-x2",
    "model-chai-2"
   ]
  },
  "reading_notes": [
   "The same campaign has two denominators (release page nine targets, manuscript ten); read them by version and never merge them into \"6/9 in both modalities\".",
   "Rates are per target, not per design: 60,000 designs per target, only the top 15 assayed.",
   "Wet-lab work was CRO- and partner-executed and developer-compiled, so it is not independent replication; the \"26 external labs\" framing has no source.",
   "No site benchmark entry exists, so comparability rests on the wet lab and on the same easy-target set used by Chai-2."
  ],
  "caveats": [
   "Zero pairings: the manuscript reports no site benchmark.",
   "Release page and manuscript disagree on target and hit counts; listed per version.",
   "Partner campaign target counts and success rates are not given as numbers (unverified).",
   "The successor BoltzProt-1 is closed and its claimed gain is reported two conflicting ways."
  ],
  "sources": [
   {
    "url": "withheld: author-hosted manuscript",
    "title": "BoltzGen manuscript PDF",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12697729/",
    "title": "BoltzGen preprint record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://jclinic.mit.edu/boltzgen/",
    "title": "Research-institute release page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/HannesStark/boltzgen/releases",
    "title": "GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.tamarind.bio/blog/boltzmol-1-boltzprot-1-and-the-boltz-api-are-live-on-tamarind-on-day-0",
    "title": "BoltzProt-1 note reposted by a third-party platform (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "6/9 for both modalities",
    "now": "Manuscript: nanobody 6/10, miniprotein 5/10; the release page separately records 6/9 (nanobody); listed separately by version",
    "source_url": "withheld: author-hosted manuscript",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Easy targets 4/5",
    "now": "Nanobody 4/5, miniprotein 3/5",
    "source_url": "withheld: author-hosted manuscript",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "26 targets validated by 26 external labs",
    "now": "No number given: the manuscript does not say \"external labs\"; a CRO ran the work",
    "source_url": "withheld: author-hosted manuscript",
    "date_found": "2026-11-16"
   },
   {
    "field": "release_milestones.paper_v1",
    "was": "No bioRxiv entry",
    "now": "bioRxiv v1 exists (2025-11-24), not published in a journal",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12697729/",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2025-11",
    "now": "2025-10-26 (code and weights public)",
    "source_url": "https://github.com/HannesStark/boltzgen",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[4]",
   "background.funding",
   "background.products_partnerships",
   "first_release",
   "open_status.code_license",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-boltzgen/",
   "en": "/bio/benchmark/models/boltzgen/"
  },
  "names_full_null_reason": "No expanded name; the release page uses BoltzGen only",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-borzoi",
  "kind": "model",
  "short_slug": "borzoi",
  "names": {
   "display": "Borzoi",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "Borzoi"
   ],
   "ambiguous_aliases": [
    "Flashzoi",
    "Borzoi Prime"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "genomics-rna",
  "domains_secondary": [],
  "one_liner": "Predicts RNA-seq coverage from DNA and scores variant effects on transcription, splicing, polyadenylation",
  "org": {
   "names": [
    "Calico Life Sciences"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Sequence-to-function model",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Open source; code under Apache-2.0"
  },
  "first_release": {
   "date": "2023-09-01",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 posted (Crossref issued the same day)",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.08.30.555582",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2023-09-01",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.08.30.555582",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-01-08",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Nature Genetics publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41588-024-02053-6",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2024-10-23",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub v1.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/calico/borzoi/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-01-08",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Nature Genetics publication",
   "description": "Published in Nature Genetics; later commits to 2026-09-30 are CI and dependency maintenance only",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41588-024-02053-6",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only maintenance commits after publication",
  "what_it_does": "Predicts RNA-seq coverage directly from DNA sequence and uses it to score variant effects on transcription, splicing and polyadenylation; continues the Basenji/Enformer line and is in the same class as AlphaGenome.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "ENCODE RNA-seq, GTEx v8, FANTOM5 CAGE, DNase, ChIP-seq etc.; 7,611 human output tracks",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-08-enformer",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": null,
    "source_url": "https://doi.org/10.1038/s41588-024-02053-6",
    "confidence": "high",
    "verification": "verified",
    "note": "Developer-reported; paper uses its own eQTL/sQTL tasks and the Enformer track test"
   },
   {
    "benchmark_id": "bench-37-traitgym",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": null,
    "source_url": "https://doi.org/10.1101/2025.02.11.637758",
    "confidence": "medium",
    "verification": "partial",
    "note": "TraitGym authors evaluated Borzoi after the fact"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": "TraitGym authors find it stronger on complex non-disease traits but behind alignment-based models on Mendelian and disease traits",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Calico is an Alphabet aging-biology company announced 2013-09; Basenji and Borzoi come from its team",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "The AbbVie partnership in place since 2014 reportedly ended in 2025-11 (press report)",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphagenome",
    "model-ntv3"
   ]
  },
  "reading_notes": [
   "Enformer test-set comparisons were run by the developers",
   "TraitGym results show clear differences across trait classes"
  ],
  "caveats": [
   "bioRxiv posting day 2023-08-31 vs 09-01 is uncertain; API record gives 09-01",
   "Borzoi Prime and Flashzoi not verified and not merged here"
  ],
  "sources": [
   {
    "url": "https://api.crossref.org/works/10.1038/s41588-024-02053-6",
    "title": "Crossref record 10.1038/s41588-024-02053-6",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1101/2023.08.30.555582",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/calico/borzoi/releases",
    "title": "GitHub repository calico/borzoi",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41588-024-02053-6",
    "title": "Journal article DOI 10.1038/s41588-024-02053-6",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2023.08.30.555582",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2023-08-31",
    "now": "2023-09-01 (bioRxiv v1 and Crossref issued date)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.08.30.555582",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-borzoi/",
   "en": "/bio/benchmark/models/borzoi/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-bulkformer",
  "kind": "model",
  "short_slug": "bulkformer",
  "names": {
   "display": "BulkFormer",
   "full": null,
   "aliases": [
    "BulkFormer",
    "BulkFormer-Atlas",
    "BulkFormer-Latest"
   ],
   "ambiguous_aliases": []
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "BulkFormer (paper checkpoints)",
     "id_or_null": null,
     "first_release": {
      "date": "2025-06-17",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint v1",
      "event_kind": "paper-revision",
      "timestamp_utc": null,
      "source_url": "https://www.biorxiv.org/content/10.1101/2025.06.11.659222.full.pdf",
      "verification": "unverified",
      "note": null
     }
    },
    {
     "name": "BulkFormer-Atlas",
     "id_or_null": null,
     "first_release": {
      "date": "2026-11-04",
      "precision": "day",
      "year_hint": null,
      "basis": "Repository tag release",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://github.com/KangBoming/BulkFormer",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "BulkFormer-Latest",
     "id_or_null": null,
     "first_release": {
      "date": "2026-11-06",
      "precision": "day",
      "year_hint": null,
      "basis": "Final checkpoint committed",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://github.com/KangBoming/BulkFormer",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "The paper version spans five sizes, with two larger variants added in November 2026 under different hosting and licence statements"
  },
  "family_null_reason": null,
  "domain": "genomics-rna",
  "domains_secondary": [
   "single-cell"
  ],
  "one_liner": "Foundation model for human bulk transcriptomes at about 150 million parameters",
  "org": {
   "names": [
    "Peking University (biomedical informatics)",
    "Jilin University",
    "Wuhan Sports University"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Transcriptome foundation model (masked reconstruction)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Paper-version code and weights under MIT; the two November 2026 variants are hosted on a general file-sharing service with no separate licence statement"
  },
  "first_release": {
   "date": "2025-06-17",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://www.biorxiv.org/content/10.1101/2025.06.11.659222.full.pdf",
   "verification": "unverified",
   "note": "The date comes from the existing record and was not rechecked this round"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-06-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.06.11.659222.full.pdf",
    "verification": "unverified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-06-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.06.11.659222.full.pdf",
    "verification": "unverified",
    "note": null
   },
   "journal_online": {
    "date": "2026-07",
    "precision": "month",
    "year_hint": null,
    "basis": "Journal publication (registry records that month)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1016/j.cels.2026.101657",
    "verification": "partial",
    "note": "A Chinese report gives 1 July 2026 online, not reproduced on the publisher page"
   },
   "code_release": {
    "date": "2026-11-04",
    "precision": "day",
    "year_hint": null,
    "basis": "Repository tag releasing two new variants",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://github.com/KangBoming/BulkFormer",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-11-06",
    "precision": "day",
    "year_hint": null,
    "basis": "Final checkpoint committed",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://github.com/KangBoming/BulkFormer",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-11-06",
   "precision": "day",
   "year_hint": null,
   "basis": "Final checkpoint committed",
   "event_kind": "new-weights",
   "timestamp_utc": null,
   "source_url": "https://github.com/KangBoming/BulkFormer",
   "verification": "verified",
   "description": "The repository tagged two new model variants on 4 November 2026 and committed the final checkpoint of one on 6 November 2026; the journal paper appeared in July 2026.",
   "note": "The site recorded 2026-11-04; the later commit is 11-06"
  },
  "status": "active",
  "status_note": "Journal publication and new weights within 12 months",
  "what_it_does": "Works on human bulk RNA sequencing: it takes expression over about twenty thousand protein-coding genes, pretrains by masking and reconstructing values with a gene co-expression graph as an attention prior, and attaches simple heads to gene or sample embeddings for imputation, disease annotation, prognosis, drug response, compound perturbation and gene essentiality. The other omics foundation models here, such as CellFM, scFoundation and scGPT/Geneformer, pretrain on single-cell data; this is the only bulk one, and those models are its comparisons.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Human bulk transcriptomes from two public expression archives covering 20,010 protein-coding genes; the preprint gives 522,769 samples and the published abstract 581,503",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.06.11.659222.full.pdf"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Downstream tasks use public cohorts and public drug-response and essentiality datasets",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Self-built downstream task set",
    "category": "own-task-set",
    "detail": "The preprint abstract lists six downstream tasks and the published abstract five, against single-cell foundation models and simple baselines",
    "source_type": "self"
   },
   {
    "label": "Third-party pan-cancer mutation probing",
    "category": "public-benchmark-not-on-site",
    "detail": "A corporate lab's conference abstract probes frozen embeddings linearly and places this model behind two single-cell foundation models",
    "source_type": "third"
   },
   {
    "label": "A snapshot of a public bulk leaderboard",
    "category": "public-benchmark-not-on-site",
    "detail": "The repository states that the new variant combined with the organiser's baseline topped one day's snapshot while ranking lower alone, and warns that rankings move",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a roughly 150-million-parameter bulk transcriptome foundation model pretrained on over half a million public samples that beats single-cell foundation models and simple baselines across its chosen downstream tasks.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A corporate lab's conference abstract ranks it behind two single-cell foundation models on pan-cancer mutation probing, and a research institute's preprint uses it as a comparison reference.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "A biomedical informatics lab in a Chinese university's school of basic medical sciences together with a national key laboratory, with co-authors from two other universities.",
   "funding": "The preprint acknowledges one national natural science foundation grant and the published version lists an additional grant number.",
   "products_partnerships": "Code and weights are maintained in a public repository and model hub, with the repository also recording a submission to a public bulk leaderboard; no company or pharmaceutical partnership is disclosed.",
   "related_models": [
    "model-cellfm",
    "model-xtrimoscperturb-scfoundation",
    "model-scgpt-geneformer"
   ]
  },
  "reading_notes": [
   "The sweep is over tasks the authors chose, comparing models pretrained for single-cell data through one shared interface.",
   "A third-party comparison places it behind two single-cell models, so the advantage varies by task rather than covering bulk work generally.",
   "Preprint and published version differ on sample count and task count, so citations must state the version.",
   "Its value is filling the gap left by single-cell-only coverage; by foundation-model standards it is small."
  ],
  "caveats": [
   "It has no pairing among the 44 benchmarks here; all evaluations are self-built.",
   "The first-release date comes from the existing record and was not rechecked.",
   "The journal date is verifiable only to the month; the day given in a Chinese report was not reproduced.",
   "The November 2026 weights sit on a general file-sharing service with no separate licence statement.",
   "The leaderboard snapshot is self-reported, and the developer itself warns the ranking moves."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1016/j.cels.2026.101657",
    "title": "Journal paper 2026",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.06.11.659222.full.pdf",
    "title": "Preprint 2025",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/KangBoming/BulkFormer",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.lgresearch.ai/publication/view?seq=157",
    "title": "Third-party conference abstract 2026",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-11-04 (new release tag)",
    "now": "2026-11-06 (final checkpoint commit)",
    "source_url": "https://github.com/KangBoming/BulkFormer",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "release_milestones.journal_online",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-bulkformer/",
   "en": "/bio/benchmark/models/bulkformer/"
  },
  "names_full_null_reason": "The name is a compound word with no expanded form in the paper",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-c2s-scale",
  "kind": "model",
  "short_slug": "c2s-scale",
  "names": {
   "display": "C2S-Scale",
   "full": "Cell2Sentence-Scale",
   "aliases": [
    "C2S-Scale",
    "Cell2Sentence"
   ],
   "ambiguous_aliases": []
  },
  "family": "Cell2Sentence",
  "family_null_reason": null,
  "domain": "single-cell",
  "domains_secondary": [
   "text-agents"
  ],
  "one_liner": "Single-cell model that turns ranked expression into \"cell sentences\" and fine-tunes general LLMs from 410M to 27B parameters",
  "org": {
   "names": [
    "Yale University",
    "Google Research",
    "Google DeepMind"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Single-cell language model",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "CC-BY-4.0",
   "code_license": null,
   "license_changed": null,
   "note": "Code is in a public repository and weights on the model hub with a Creative Commons licence on the model card; the code licence was not re-checked this round"
  },
  "first_release": {
   "date": "2025-04-17",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1",
   "event_kind": "preprint-posted",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
   "verification": "verified",
   "note": "The site recorded 2025-04, month level only"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-04-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-04-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2025-10",
    "precision": "month",
    "year_hint": null,
    "basis": "Largest-size weights released on the model hub alongside an official blog post",
    "event_kind": "weights-release",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/vandijklab/C2S-Scale-Gemma-2-27B",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-01-01",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v4",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
   "verification": "verified",
   "description": "The preprint was revised to v4 (v2 on 2025-10-11, v3 on 2025-10-26); the earlier event is the 2025-10 release of the largest size. No journal version was found",
   "note": null
  },
  "status": "active",
  "status_note": "Preprint revisions and a weights release within 12 months",
  "what_it_does": "Ranks each cell's most expressed genes into a text sequence and uses a general language model for cell-type annotation, conditional generation and question answering, extending to spatial neighbourhoods and perturbation response; it keeps the general LLM architecture rather than a single-cell-specific one, the opposite of TranscriptFormer. The developers report that some context-dependent reasoning appears only at the largest size.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Over 50 million human and mouse cells from public cell atlases and other sources plus biological text and metadata, about 1 billion tokens and 150 million training samples",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.04.14.648850v1.full"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.04.14.648850v1.full"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.04.14.648850v1.full"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "One drug hypothesis was tested in the lab, as validation rather than training",
    "source_url": "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Experimental test of a single drug hypothesis",
    "targets": 1,
    "designs_tested": 1,
    "hits": 1,
    "denominator_basis": "One hypothesis taken forward from a computational screen of over 4,000 drugs; batch and replicate counts are not public, so no hit rate can be computed",
    "success_rate": "1/1 hypothesis tested",
    "affinity_best": null,
    "version": "2025-10 and later preprint versions",
    "source_type": "self",
    "source_url": "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/",
    "date": {
     "date": "2025-10",
     "precision": "month",
     "year_hint": null,
     "basis": "Official blog and contemporaneous preprint version",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/",
     "verification": "partial",
     "note": null
    },
    "verification": "partial",
    "executor": "Run by the collaborating academic lab; not an independent reproduction",
    "unverified_marker": "Replicate counts and executor attribution are unverified; no replicate figure is given",
    "note": "The model predicted that a kinase inhibitor would raise antigen presentation only under low-interferon conditions, and an increase was seen in a previously unseen human cell model; the result exists only in the 2025-10 and later preprint versions, so citations must point at the current version"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://www.biorxiv.org/content/10.1101/2025.04.14.648850v1.full",
    "confidence": "high",
    "second_hand": false,
    "note": "Perturbation prediction uses self-chosen datasets and self-defined metrics against three comparable methods; it is not that evaluation's official track"
   }
  ],
  "other_evaluations": [
   {
    "label": "Cell-type annotation comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "Self-reported parity with two single-cell foundation models on annotation; a comparison rather than a public-protocol evaluation",
    "source_type": "self"
   },
   {
    "label": "Self-chosen metric for integration",
    "category": "own-task-set",
    "detail": "The preprint does not use the common integration metrics and substitutes another alignment measure, which cannot be listed beside the site's integration evaluation",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that handling single-cell data with a general language model improves with scale and can produce biological hypotheses testable in the lab.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The preprint is not peer reviewed and no third-party evaluation of its perturbation or annotation results was found",
  "background": {
   "org_profile": "A collaboration between a US university lab and two research units of a large technology company, built on that company's open model family.",
   "funding": "The corporate partner supplied models and compute; no grant numbers are listed on the paper page or the blog.",
   "products_partnerships": "The model family was released in 2025-04 and the largest size in 2025-10 with an official blog post; the same company's genomics work is AlphaGenome.",
   "related_models": [
    "model-transcriptformer",
    "model-alphagenome",
    "model-scgpt-geneformer"
   ]
  },
  "reading_notes": [
   "Comparisons are self-reported, with the main text giving directional conclusions and per-item values left to the supplement.",
   "The drug-hypothesis test is a single case, not systematic wet-lab backing.",
   "It reports no public integration protocol, so it cannot be compared directly with single-cell models that do.",
   "The wet-lab result appears only from the 2025-10 version onward, so citing the earlier version misses it"
  ],
  "caveats": [
   "The site linked the first preprint version, which does not contain the wet-lab result; the link must point at the current version.",
   "Replicate counts and executor attribution are unverified; no replicate figure is given.",
   "The preprint is not peer reviewed, so all comparison values are the authors' own."
  ],
  "sources": [
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.04.14.648850v1.full",
    "title": "Preprint full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
    "title": "Preprint version record",
    "kind": "registry",
    "date": null
   },
   {
    "url": "https://huggingface.co/vandijklab/C2S-Scale-Gemma-2-27B",
    "title": "Model hub weights page",
    "kind": "model-hub",
    "date": null
   },
   {
    "url": "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/",
    "title": "Official blog (wet-lab result)",
    "kind": "blog",
    "date": null
   },
   {
    "url": "https://github.com/vandijklab/cell2sentence",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-04",
    "now": "2025-04-17 (preprint v1 posted)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2025-10-15 (weights release, recorded as new weights)",
    "now": "2026-01-01 (preprint v4, kind paper revision); the weights release is recorded as an earlier event",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.04.14.648850",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Cited the first preprint version",
    "now": "Cites the 2025-10 and later versions; the first version does not contain this result",
    "source_url": "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.weights_release",
   "wetlab_validation[0]",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-c2s-scale/",
   "en": "/bio/benchmark/models/c2s-scale/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-cellfm",
  "kind": "model",
  "short_slug": "cellfm",
  "names": {
   "display": "CellFM",
   "full": null,
   "aliases": [
    "CellFM"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single model with no family of versions",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Human-only single-cell foundation model with 800 million parameters",
  "org": {
   "names": [
    "Sun Yat-sen University (computer science)",
    "Chongqing University",
    "Huawei",
    "Singleron Biotechnologies"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Single-cell foundation model (linear-complexity backbone)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code in a public repository with a long-term archive and weights in a public model hub; licence text not retrieved this round"
  },
  "first_release": {
   "date": "2024-06-06",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint posted (registry record)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1038/s41467-025-59926-5",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-06-06",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41467-025-59926-5",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-06-06",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.06.04.597369.full.pdf",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-05-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal publication",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41467-025-59926-5",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-09-07",
   "precision": "day",
   "year_hint": null,
   "basis": "Author correction statement in the repository README",
   "event_kind": "author-correction",
   "timestamp_utc": null,
   "source_url": "https://github.com/biomed-AI/CellFM",
   "verification": "unverified",
   "description": "An author correction in the repository README: miscalculated averages in several figure panels plus evaluation bugs affecting two baselines on two datasets, with the authors stating the conclusions are unchanged. The earlier event is the journal publication on 20 May 2025.",
   "note": "No independent permalink for the commit day was obtained, so the date is unverified, and no formal journal erratum was found"
  },
  "status": "active",
  "status_note": "An author correction statement within 12 months",
  "what_it_does": "Pretrained on human single-cell data only, so that mixing species does not drag down performance on human cells, with a linear-complexity sequence backbone and downstream tasks spanning cell-type annotation, perturbation prediction, gene function prediction and gene relationship capture; training ran on a domestic AI framework and accelerators. It follows the same route as scGPT/Geneformer, distinguished by human-only data and domestic compute.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "19,914 samples and about 102.3 million human cells from public databases",
    "source_url": "https://doi.org/10.1038/s41467-025-59926-5"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://www.nature.com/articles/s41467-025-59926-5",
    "confidence": "high",
    "second_hand": false,
    "note": "Uses two early public perturbation datasets with one shared downstream predictor rather than the challenge's blind test, and reports gains as relative percentages"
   }
  ],
  "other_evaluations": [
   {
    "label": "Cell annotation and gene function prediction",
    "category": "public-benchmark-not-on-site",
    "detail": "Compared with two single-cell foundation models on public datasets, reporting relative gains in mean accuracy",
    "source_type": "self"
   },
   {
    "label": "Third-party perturbation evaluations",
    "category": "public-benchmark-not-on-site",
    "detail": "Two independent evaluations find simple baselines competitive with such foundation models; neither includes this model",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: human-only pretraining on about a hundred million cells with a linear-complexity backbone, beating comparable models on annotation, perturbation prediction and gene function prediction, mostly as single-digit relative gains.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Two third-party evaluations do not include this model, and no independent evaluation of it was found",
  "background": {
   "org_profile": "An academic team in a Chinese university's computer science school, with collaborators at another university, a laboratory, a telecommunications and compute company, an affiliated hospital and a single-cell sequencing company.",
   "funding": "The acknowledgements list national key R&D programme, national natural science foundation, central university research funds, postdoctoral funds and a provincial natural science foundation.",
   "products_partnerships": "Code sits under a public repository organisation with a weights card in a public model hub; comparable models include scFoundation and scGPT/Geneformer.",
   "related_models": [
    "model-scgpt-geneformer",
    "model-xtrimoscperturb-scfoundation",
    "model-state"
   ]
  },
  "reading_notes": [
   "All performance claims are self-reported as relative gains over the runner-up, so the absolute gaps are small.",
   "Independent work finds simple baselines competitive with such foundation models on perturbation prediction, and that doubt applies here too.",
   "The authors posted their own correction of figure and evaluation bugs in the repository, which is commendable, though no formal journal erratum was found.",
   "Its value lies more in human-only data and the feasibility of domestic compute than in a verified predictive edge."
  ],
  "caveats": [
   "No independent permalink was obtained for the correction commit's day.",
   "No formal journal erratum was found; the correction appears only in the repository README.",
   "It has no result on the public blind-test challenge.",
   "Licence text was not retrieved this round.",
   "The perturbation evaluation uses early public datasets, which differ from current blind-test protocols."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41467-025-59926-5",
    "title": "Journal paper 2025",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2024.06.04.597369.full.pdf",
    "title": "Preprint 2024",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/biomed-AI/CellFM",
    "title": "Code repository and correction statement",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12328236/",
    "title": "Third-party perturbation evaluation",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2025-05-20 (journal publication)",
    "now": "2026-09-07 (author correction notice in the repository); event kind changed to author correction",
    "source_url": "https://github.com/biomed-AI/CellFM",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "open_status.code_license",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-cellfm/",
   "en": "/bio/benchmark/models/cellfm/"
  },
  "names_full_null_reason": "The name is an acronym compound with no expanded form in the paper",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-chai-1",
  "kind": "model",
  "short_slug": "chai-1",
  "names": {
   "display": "Chai-1",
   "full": null,
   "full_null_reason": "Official materials use Chai-1 only; no expanded name",
   "aliases": [
    "Chai-1",
    "chai-lab"
   ],
   "ambiguous_aliases": [
    "Chai"
   ]
  },
  "family": null,
  "family_null_reason": "Covers Chai-1 only; Chai-2 and Chai-3 have their own records",
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "Open-weights co-folding model that stays stable without MSAs and accepts experimental restraints",
  "org": {
   "names": [
    "Chai Discovery"
   ],
   "global_entity_ids": [
    "chai-discovery"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding (protein / nucleic acid / small molecule / glycan)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0 (from v0.4.0, 2024-11-27)",
   "code_license": "Apache-2.0 (from v0.4.0, 2024-11-27)",
   "license_changed": {
    "date": "2024-11-27",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "From chai-lab v0.4.0 both code and weights are Apache-2.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/chaidiscovery/chai-lab",
    "verification": "partial",
    "note": null
   },
   "note": "Code and weights share one licence allowing commercial use; the initial terms before v0.4.0 were not re-verified this round"
  },
  "first_release": {
   "date": "2024-09-09",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Official release post with the repository and web server at the same time; preprint 2024-10-11",
   "timestamp_utc": null,
   "source_url": "https://www.chaidiscovery.com/news/introducing-chai-1",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-10-11",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2024.10.10.615955",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-09-09",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Repository made public",
    "timestamp_utc": null,
    "source_url": "https://github.com/chaidiscovery/chai-lab",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2024-09-09",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights provided at launch",
    "timestamp_utc": null,
    "source_url": "https://github.com/chaidiscovery/chai-lab",
    "verification": "partial",
    "note": "Initial terms not re-verified; Apache-2.0 from v0.4.0 (2024-11-27)"
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-09-09",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Official release post",
    "timestamp_utc": null,
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-1",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-06-04",
   "precision": "day",
   "event_kind": "successor-model",
   "basis": "The successor was first disclosed through a licence announcement",
   "description": "Chai-3 was first disclosed as a partner-only successor with no public weights; the open repository's last release was 2025-03-18",
   "timestamp_utc": null,
   "source_url": "https://www.chaidiscovery.com/news",
   "verification": "partial",
   "note": "Outlets report dates between 06-02 and 06-05 for this event (partially verified)"
  },
  "status": "superseded",
  "status_note": "The company moved to successor design models; no open release since 2025-03",
  "what_it_does": "It predicts protein, nucleic-acid, small-molecule and glycan complexes. Its distinctive mode is single-sequence inference: accuracy drops only modestly without MSAs, which suits novel targets and antibodies, and experimental restraints can be supplied. It is one open-weights alternative to AlphaFold 3, and splits prediction from design with its successor Chai-2.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB and other public sources; no cutoff date or entry counts given officially",
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-1"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No wet-lab work reported",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "protein monomers of one past season",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): a second-hand quote only, and post hoc"
   },
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "PoseBusters V1 (428), official set",
    "source_url": "https://doi.org/10.1101/2024.10.10.615955",
    "confidence": "high",
    "verification": "partial",
    "note": "The self-report is quoted via reviews and the full text was not read; stricter-protocol figures are third-party re-runs"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "Runs N' Poses, best@5",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): quoted from a third-party paper"
   },
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer leaderboard",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "high",
    "verification": "verified",
    "note": null
   },
   {
    "benchmark_id": "bench-33-posex",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "benchmark authors' own sets",
    "source_url": "https://arxiv.org/html/2505.01700v3",
    "confidence": "medium",
    "verification": "partial",
    "note": null
   },
   {
    "benchmark_id": "bench-35-flab",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "zero-shot thermostability eval in the benchmark's own paper",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12767642/",
    "confidence": "medium",
    "verification": "partial",
    "note": "Evaluated in the benchmark's own paper; the site matrix calls it third-party while the benchmark page calls it a self-report, and the labels are still to be unified"
   }
  ],
  "other_evaluations": [
   {
    "label": "Third-party MSA-free robustness evaluation",
    "category": "third-party-evaluation",
    "detail": "A third-party evaluation concludes it is the only method that stays stable without MSAs",
    "source_type": "third"
   },
   {
    "label": "Third-party tables outside the maintainer board",
    "category": "third-party-evaluation",
    "detail": "Several successor-model papers include it in comparison tables, all third-party re-runs under differing protocols",
    "source_type": "third"
   },
   {
    "label": "Self-reported results quoted via reviews",
    "category": "comparison-vs-baseline-only",
    "detail": "The preprint's self-reported results were read only through reviews this round, not from the full text",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: co-folding accuracy on par with the contemporary closed model, little loss without MSAs, and conditioning on experimental restraints.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Third-party re-runs under stricter protocols place it below its self-reported level; one third-party evaluation independently confirms its MSA-free robustness, and a maintainer leaderboard includes it.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The company was founded in 2024-03 in San Francisco, worked early on from an AI lab's offices, and positions itself as an AI-native molecular design platform selling models and custom models rather than running its own pipeline; headcount undisclosed.",
   "funding": "A $30M seed in 2024-09, a $70M Series A in 2025-08, a $130M Series B at $1.3B on 2025-12-15 and a $400M Series C at $3.8B in 2026-07.",
   "funding_null_reason": null,
   "products_partnerships": "The product line runs from this open-weights model to Chai-2 for design and partner-only Chai-3; collaborations with four large pharma companies include custom models and design kits, all with undisclosed terms.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-chai-2",
    "model-chai-3",
    "model-alphafold3",
    "model-boltz-1-1x",
    "model-boltz-2",
    "model-protenix"
   ]
  },
  "reading_notes": [
   "The self-report uses the looser official protocol while third-party re-runs under stricter rules land much lower, so the self-report should not be read as beating the contemporary closed model.",
   "Its MSA-free robustness comes from a third-party evaluation, which is firmer than the self-report and is its most distinctive finding.",
   "Training-data information is minimal, with no cutoff or entry counts, so leakage risk cannot be assessed.",
   "The antibody-property pairing carries inconsistent source labels in two places on the site; reading it as the benchmark side's own paper is safer."
  ],
  "caveats": [
   "The self-reported results were read only via reviews, not the original full text (partially verified).",
   "Two pairings rest only on the site's own pages (unverified).",
   "Outlets differ by a few days on the successor's disclosure date (partially verified).",
   "The initial weights terms were not re-verified; only the Apache-2.0 status from v0.4.0 is confirmed."
  ],
  "sources": [
   {
    "url": "https://www.chaidiscovery.com/news/introducing-chai-1",
    "title": "Official release post",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2024.10.10.615955",
    "title": "The model's preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/chaidiscovery/chai-lab",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2510.24670",
    "title": "Third-party re-run paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2505.01700v3",
    "title": "Re-run paper by the benchmark's authors",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12767642/",
    "title": "Paper from the antibody-property benchmark side",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://github.com/BEAM-Labs/FoldBench",
    "title": "Maintainer leaderboard repository",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://www.chaidiscovery.com/news",
    "title": "Official news list (successor disclosure)",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "The first release was given as a 2024-09/10 range",
    "now": "Changed to the day of the official release post, with the preprint listed as a separate milestone",
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-1",
    "date_found": "2026-11-16"
   },
   {
    "field": "open_status",
    "was": "The licence was described as Apache-2.0 from launch",
    "now": "Changed to Apache-2.0 from v0.4.0 (2024-11-27), with the initial terms not re-verified",
    "source_url": "https://github.com/chaidiscovery/chai-lab",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "The antibody-property row carried both third-party and self-report labels",
    "now": "Unified as retrospective dataset use in the benchmark side's own paper, with the label mismatch recorded",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12767642/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[0]",
   "benchmarks_reported[2]",
   "latest_update",
   "release_milestones.weights_release",
   "training_data.b_private_licensed"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-chai-1/",
   "en": "/bio/benchmark/models/chai-1/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-chai-2",
  "kind": "model",
  "short_slug": "chai-2",
  "names": {
   "display": "Chai-2",
   "full": null,
   "full_null_reason": "Official materials use Chai-2; the built-in folding module is called Chai-2f",
   "aliases": [
    "Chai-2",
    "Chai-2f"
   ],
   "ambiguous_aliases": [
    "Chai"
   ]
  },
  "family": null,
  "family_null_reason": "Covers Chai-2 and its folding module only; Chai-1 and Chai-3 have their own records",
  "domain": "structure",
  "domains_secondary": [
   "antibody",
   "generation"
  ],
  "one_liner": "Closed zero-shot antibody and binder design model with an upgraded folding module built in",
  "org": {
   "names": [
    "Chai Discovery"
   ],
   "global_entity_ids": [
    "chai-discovery"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (antibody / nanobody / miniprotein) plus built-in folding",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Closed and partner-only, with no public weights or inference code"
  },
  "first_release": {
   "date": "2025-06-30",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Official research post and press release the same day",
   "timestamp_utc": null,
   "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-06-30",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Official research post",
    "timestamp_utc": null,
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-06-04",
   "precision": "day",
   "event_kind": "successor-model",
   "basis": "The successor was disclosed through a licence announcement",
   "description": "Partner-only successor Chai-3 was first disclosed; an earlier post on 2025-11-19 covered full-length antibody design",
   "timestamp_utc": null,
   "source_url": "https://www.chaidiscovery.com/news",
   "verification": "partial",
   "note": "The official news page has no dedicated post for the successor (partially verified)"
  },
  "status": "product-only",
  "status_note": "Closed and partner-only with no public weights; the successor is partner-only too",
  "what_it_does": "Given a target and epitope residues it generates antibodies, nanobodies or miniproteins zero-shot without a known starting binder, with a folding module included for structural assessment. Its role differs from general co-folding in Chai-1, and BindCraft, BoltzGen and RFdiffusion2/3 are open counterparts whose hit-rate definitions differ.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "The technical report does not say whether it has own training data",
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Design validation on 52 targets; no public statement that results were fed back into training",
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Antibody and nanobody design campaign",
    "modality": "VHH; scFv",
    "targets": 52,
    "designs_tested": 20,
    "hits": 26,
    "denominator_basis": "52 novel antigens with up to 20 designs per target per format (VHH and scFv), hits called by a binding-signal threshold",
    "success_rate": "binders for 26/52 targets; ~16% mean per-design hit rate (self-reported)",
    "affinity_best": null,
    "version": "release page",
    "source_type": "self",
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2",
    "date": {
     "date": "2025-06-30",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "Published with the release page",
     "timestamp_utc": null,
     "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "note": "Self-reported on self-chosen targets with no independent replication; the per-design and per-target rates must not be mixed"
   },
   {
    "campaign": "Miniprotein wet-lab test",
    "modality": "miniprotein",
    "targets": 5,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Five targets with 20-25 designs each, read out by binding signal",
    "success_rate": "68% success (self-reported)",
    "affinity_best": null,
    "version": "release page",
    "source_type": "self",
    "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2",
    "date": {
     "date": "2025-06-30",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "Published with the release page",
     "timestamp_utc": null,
     "source_url": "https://www.chaidiscovery.com/news/introducing-chai-2",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "note": "Comparisons with other methods' published figures are citations, not re-runs; not an independent replication"
   },
   {
    "campaign": "2025-11 full-length antibody update (unverified)",
    "modality": "IgG",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Secondary reports only with no primary text located, so neither numerator nor denominator is given (unverified)",
    "success_rate": null,
    "affinity_best": null,
    "version": "2025-11 update",
    "source_type": "self",
    "source_url": "https://www.chaidiscovery.com/news",
    "date": {
     "date": "2025-11-19",
     "precision": "month",
     "event_kind": "product-update",
     "basis": "Developer update post (relayed second-hand)",
     "timestamp_utc": null,
     "source_url": "https://www.chaidiscovery.com/news",
     "verification": "unverified",
     "note": null
    },
    "verification": "unverified",
    "note": "The two ratios published earlier could not be matched to any primary source, so no figures are given under this round's rules (unverified)"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Antibody and nanobody wet-lab campaign",
    "category": "wet-lab-validation",
    "detail": "52 novel antigens with up to 20 designs per target per format, hits by binding signal, compared with earlier de novo antibody methods",
    "source_type": "self"
   },
   {
    "label": "Miniprotein wet-lab test",
    "category": "wet-lab-validation",
    "detail": "Five targets with 20-25 designs each, compared with the published results of two earlier methods",
    "source_type": "self"
   },
   {
    "label": "Antibody-antigen folding module self-evaluation",
    "category": "comparison-vs-baseline-only",
    "detail": "On 152 low-homology antibody-antigen complexes (255 interfaces) against the predecessor and two baselines; scores are not shown here",
    "source_type": "self"
   },
   {
    "label": "Developability and polyreactivity own-task set",
    "category": "own-task-set",
    "detail": "Polyreactivity assays on the designs plus developability and immunogenicity scores against approved antibodies and paired natural sequences",
    "source_type": "self"
   },
   {
    "label": "2025-11 full-length antibody update",
    "category": "user-or-partner-report",
    "detail": "A developer update on full-length antibody developability, known only through secondary reports; no figures given (unverified)",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: zero-shot double-digit per-design hit rates from small design sets, binders for about half of the novel antigens, and antibody-antigen docking accuracy better than the predecessor.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A competitor report (JAM/JAM-2) re-tested 16 overlapping targets and compares bind rates; no neutral third-party replication was found.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The company was founded in 2024-03 in San Francisco, worked early on from an AI lab's offices, and positions itself as an AI-native molecular design platform selling models and custom models rather than running its own pipeline; headcount undisclosed.",
   "funding": "About $30M seed in 2024-09, a $70M Series A on 2025-08-06, a $130M Series B at $1.3B on 2025-12-15 and a $400M Series C at $3.8B in 2026-07; the company gives no official cumulative figure.",
   "funding_null_reason": null,
   "products_partnerships": "The line runs from open-weights Chai-1 through this design model to partner-only Chai-3; four pharma collaborations through 2026 are press releases with no verifiable performance data and undisclosed terms.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-chai-1",
    "model-chai-3",
    "model-bindcraft",
    "model-boltzgen",
    "model-latent-x2",
    "model-jam-2"
   ]
  },
  "reading_notes": [
   "Every figure is a developer self-report on self-chosen targets, with no third-party replication and no result in a common public challenge.",
   "Its hit-rate definitions (per design or per target, binding or function) differ from other design models, so cross-model ranking has no basis.",
   "It is closed and discloses no training data, so overlap between training and test targets cannot be assessed.",
   "The collaboration announcements are press releases covering access only, and the successor has no public benchmark at all."
  ],
  "caveats": [
   "Zero pairings: this record has no site benchmark entry, so comparability rests on self-reported wet lab.",
   "The 2025-11 full-length antibody update is treated as unverified and carries no figures.",
   "Only a DOI-string date exists for the preprint v1, which this round excludes, so that milestone is left empty.",
   "Second-hand paraphrases of the successor doubling success rates are inconsistent and are not adopted here.",
   "The wet lab is developer-reported with formats and targets of its own choosing, not an independent evaluation."
  ],
  "sources": [
   {
    "url": "https://www.chaidiscovery.com/news/introducing-chai-2",
    "title": "Official research post (release page)",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.07.05.663018",
    "title": "Companion technical report preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.chaidiscovery.com/news",
    "title": "Official news list",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.biospace.com/press-releases/chai-discovery-announces-collaboration-with-roche-in-ai-driven-drug-discovery",
    "title": "Most recent pharma collaboration press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://techcrunch.com/2025/12/15/openai-backed-biotech-firm-chai-discovery-raises-130m-series-b-at-1-3b-valuation/",
    "title": "Series B coverage (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "The full-length antibody update gave two concrete ratios",
    "now": "No figures given: they could not be matched to any primary source (unverified)",
    "source_url": "https://www.chaidiscovery.com/news",
    "date_found": "2026-11-16"
   },
   {
    "field": "other_evaluations",
    "was": "The folding self-evaluation and the design hit rates were read together",
    "now": "Separated by category: wet-lab validation, comparison versus baseline only, own task set and partner report each stand alone",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.07.05.663018",
    "date_found": "2026-11-16"
   },
   {
    "field": "background.products_partnerships",
    "was": "Paraphrased performance claims for the successor were treated as fact",
    "now": "Changed to: the collaboration announcements are press releases with no verifiable performance data",
    "source_url": "https://www.biospace.com/press-releases/chai-discovery-announces-collaboration-with-roche-in-ai-driven-drug-discovery",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[2]",
   "release_milestones.paper_v1",
   "latest_update",
   "training_data.a_public",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-chai-2/",
   "en": "/bio/benchmark/models/chai-2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-chai-3",
  "kind": "model",
  "short_slug": "chai-3",
  "names": {
   "display": "Chai-3",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "Chai-3"
   ],
   "ambiguous_aliases": [
    "Chai"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "protein",
  "domains_secondary": [
   "antibody"
  ],
  "one_liner": "Chai Discovery's closed next-generation molecular design model for antibodies and multispecifics",
  "org": {
   "names": [
    "Chai Discovery"
   ],
   "global_entity_ids": [
    "chai-discovery"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Closed molecular design model",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Closed; early access for partner pharma companies only"
  },
  "first_release": {
   "date": "2026-06-04",
   "precision": "day",
   "event_kind": "official-announcement",
   "basis": "Licence announcement on the company news page, the first public disclosure of Chai-3; no standalone technical report",
   "timestamp_utc": null,
   "source_url": "https://www.chaidiscovery.com/news",
   "verification": "partial",
   "note": "A Business Wire URL id suggests 2026-06-02, which is not a publication date"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-06-04",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Licence announcement on the company news page",
    "timestamp_utc": null,
    "source_url": "https://www.chaidiscovery.com/news",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-07-14",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Series C announcement calling Chai-3 the latest model (company post)",
   "description": "A company funding announcement calls it the latest model; a company event, not a model version",
   "timestamp_utc": null,
   "source_url": "https://www.businesswire.com/news/home/20260713849009/en/Chai-Discovery-Announces-400M-Series-C-to-Advance-AI-Driven-Molecular-Design",
   "verification": "verified",
   "note": "The 2026-07-13 collaboration and later pharma deals are not counted as model updates"
  },
  "status": "product-only",
  "status_note": "No public artifact; only partnerships and company announcements",
  "what_it_does": "Closed successor to Chai-2; the company positions it for antibody and multispecific design and harder targets. No technical report, no public benchmark, no public data.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "One pharma deal includes a custom model trained on the partner's proprietary data",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "not-disclosed",
    "detail": "The company says wet-lab validation exists; details are not public",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Company statements",
    "category": "none-disclosed",
    "detail": "Press releases claim advantages over the predecessor on binding, hard targets, multispecifics and developability; no report or data",
    "source_type": "self"
   },
   {
    "label": "Pharma partnerships",
    "category": "user-or-partner-report",
    "detail": "Several licences and collaborations announced; no partner results published",
    "source_type": "partner"
   }
  ],
  "claimed_advantage": "Developers report advantages over the predecessor on antibody design and hard targets with roughly double the success rate; no quantified public evidence",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "Chai Discovery was founded in San Francisco in 2024 as an AI-native molecular design platform",
   "funding": "Closed a $400M Series C in 2026-07 at a $3.8B valuation, led by Index Ventures",
   "funding_null_reason": null,
   "products_partnerships": "Licences or collaborations announced with several large pharma companies; financial terms undisclosed",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-chai-1",
    "model-chai-2"
   ]
  },
  "reading_notes": [
   "Press-reported success-rate figures conflict and no primary data exists",
   "Funding and partnership news are not model updates; this record follows that rule"
  ],
  "caveats": [
   "The first-release date rests on the company news page; partly verified",
   "No public benchmark results; on the zero-pairing list under this round's rules"
  ],
  "sources": [
   {
    "url": "https://www.businesswire.com/news/home/20260602498831/en/",
    "title": "Official page businesswire.com",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20260713849009/en/Chai-Discovery-Announces-400M-Series-C-to-Advance-AI-Driven-Molecular-Design",
    "title": "Official page businesswire.com",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2026-06-02",
    "now": "2026-06-04 (company news page date; a URL id is not a date source)",
    "source_url": "https://www.chaidiscovery.com/news",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "Funding and collaboration described as a model update",
    "now": "Marked as a company event, not a model version; later pharma deals excluded",
    "source_url": "https://www.chaidiscovery.com/news",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-chai-3/",
   "en": "/bio/benchmark/models/chai-3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-chemformer",
  "kind": "model",
  "short_slug": "chemformer",
  "names": {
   "display": "Chemformer",
   "full": null,
   "aliases": [
    "Chemformer"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "The organisation's multi-step retrosynthesis search tool is a separate project and is not merged into this record",
  "domain": "small-molecule",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Pretrained molecule-string sequence model for reaction prediction and retrosynthesis, long used as a baseline",
  "org": {
   "names": [
    "AstraZeneca (Molecular AI, Discovery Sciences)"
   ],
   "global_entity_ids": []
  },
  "org_type": "pharma",
  "org_type_note": null,
  "country": [
   "SE",
   "GB"
  ],
  "model_class": "Retrosynthesis / reaction prediction (string sequence model)",
  "open_status": {
   "code": "unclear",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The original repository is archived and returned unavailable through the interface this round, with no licence stated on the page; the organisation's search tool is MIT but belongs to another project"
  },
  "first_release": {
   "date": "2021-07-15",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint posted (per the registration agency record)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1088/2632-2153/ac3ffb",
   "verification": "verified",
   "note": "The site used the 2021-12 / 2022 journal dates as the release; first release follows the preprint"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2021-07-15",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1088/2632-2153/ac3ffb",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2021-07-15",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1088/2632-2153/ac3ffb",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2022-01-31",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal record issued date (the registration record's created date is 2021-12-07)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1088/2632-2153/ac3ffb",
    "verification": "partial",
    "note": "The site said online in 2021-12, which may be acceptance or record creation; both coexist"
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-02-11",
   "precision": "day",
   "year_hint": null,
   "basis": "Code repository switched to read-only archive",
   "event_kind": "repository-archived",
   "timestamp_utc": null,
   "source_url": "https://github.com/MolecularAI/Chemformer",
   "verification": "partial",
   "description": "The original code repository was archived read-only; no new version of this model has appeared since the 2022 journal paper and later development is said to have moved to another model library of the same group. This is the latest sourced event",
   "note": "Previously recorded as a code release whose description cited another project's version release from the same organisation; that event has been removed"
  },
  "status": "archived",
  "status_note": "The original repository is archived read-only and there is no new version",
  "what_it_does": "An encoder-decoder Transformer pretrained on large molecule-string corpora and fine-tuned for forward reaction prediction, retrosynthesis and molecular optimisation; its single-step predictions can be called by multi-step search tools to assemble routes. Relative to RetroDFM-R and Uni-Mol3 it serves mainly as the comparison target rather than a new method.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public patent reaction sets and public molecule-string libraries for pretraining",
    "source_url": "https://doi.org/10.1088/2632-2153/ac3ffb"
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "A company-internal reaction dataset is used for comparisons and downstream work; its scale and terms are asserted but not detailed publicly",
    "source_url": "https://arxiv.org/abs/2308.05522"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The papers report no wet-lab validation",
    "source_url": "https://doi.org/10.1088/2632-2153/ac3ffb"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-07-uspto",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.1088/2632-2153/ac3ffb",
    "confidence": "high",
    "second_hand": false,
    "note": "The official small-set retrosynthesis split as reported in the original paper; re-quotes in later models' tables count as comparison citations only and are not separate pairings. No figures are transcribed here"
   }
  ],
  "other_evaluations": [
   {
    "label": "Third-party single-step retrosynthesis benchmark",
    "category": "comparison-vs-baseline-only",
    "detail": "A 2023 third-party comparison ranks it the best template-free model and also evaluates a company-internal reaction set that outsiders cannot obtain (some figures are second-hand compilations)",
    "source_type": "third"
   },
   {
    "label": "Multi-step route success study",
    "category": "own-task-set",
    "detail": "A study from the same organisation reports higher multi-step route success than template baselines on a proprietary compound set, with a same-direction result on a public compound set; it trains on tens of millions of proprietary reactions and cannot be reproduced externally",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: after large-scale string pretraining one encoder-decoder model covers forward, reverse and optimisation sequence tasks and beats the template-free methods of the time on the official retrosynthesis split.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A 2023 third-party single-step retrosynthesis comparison rates it the best template-free model, and later model papers list it as the stronger comparator in their own tables (no figures transcribed).",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The developer is a large pharma company headquartered in the UK with three strategic R&D centres in the UK, the US and Sweden; this model comes from the molecular-AI group inside its discovery-sciences organisation in the Gothenburg area. Its 2025 annual report states revenue of about $58.7 billion and about 96,100 employees.",
   "funding": "Funded from the company's internal R&D budget with no separate financing; the papers list no external grants.",
   "products_partnerships": "The group's open-source line also includes a multi-step retrosynthesis search tool (MIT, able to call single-step models) and the molecule-generation tool REINVENT 4; these are separate projects whose releases are not events of this model. On-site, Uni-Mol3 and RetroDFM-R use it as a baseline.",
   "related_models": [
    "model-reinvent-4",
    "model-uni-mol3",
    "model-retrodfm-r"
   ]
  },
  "reading_notes": [
   "It is the steadiest sequence baseline in retrosynthesis comparisons: later models concede in their own tables that its single-step result is higher.",
   "Differences on the small USPTO split are tiny and the split leaks templates, so rankings mean little.",
   "Second-hand reports note that models trained only on patent reactions do poorly on simple medicinal-chemistry reactions, so baseline scores do not extrapolate to real chemistry.",
   "Multi-step route success rates come from proprietary company data and cannot be reproduced externally, and no new version has appeared since the repository was archived."
  ],
  "caveats": [
   "The original repository is archived and licence and weights status are unclear.",
   "The site previously recorded another project's release from the same organisation as an update to this model.",
   "The internal reaction dataset's scale and terms are asserted only.",
   "The journal online date differs between records and is marked partly verified."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1088/2632-2153/ac3ffb",
    "title": "Model journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1088/2632-2153/ac3ffb",
    "title": "Registration agency record (date basis)",
    "kind": "archive",
    "date": null
   },
   {
    "url": "https://github.com/MolecularAI/Chemformer",
    "title": "Archived code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2308.05522",
    "title": "Third-party single-step retrosynthesis comparison",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://doi.org/10.1021/acs.jcim.3c01685",
    "title": "Multi-step route study from the same organisation",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2025-12-09, a version release of another project (a multi-step search tool)",
    "now": "2026-02-11, this model's code repository archived read-only; the other project's event was removed",
    "source_url": "https://github.com/MolecularAI/Chemformer",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "code-release",
    "now": "repository-archived",
    "source_url": "https://github.com/MolecularAI/Chemformer",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2022 (journal year)",
    "now": "2021-07-15 (preprint)",
    "source_url": "https://api.crossref.org/works/10.1088/2632-2153/ac3ffb",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.journal_online",
   "latest_update",
   "open_status.code",
   "open_status.weights",
   "training_data.b_private_licensed"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-chemformer/",
   "en": "/bio/benchmark/models/chemformer/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-chemprop-v2",
  "kind": "model",
  "short_slug": "chemprop-v2",
  "names": {
   "display": "Chemprop v2",
   "full": null,
   "aliases": [
    "Chemprop v2",
    "Chemprop-RDKit",
    "Chemprop-MT"
   ],
   "ambiguous_aliases": [
    "Chemprop"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Chemprop v1 (2019 D-MPNN release)",
     "id_or_null": null,
     "first_release": {
      "date": "2019",
      "precision": "year",
      "year_hint": null,
      "basis": "First version released with the directed message-passing paper",
      "event_kind": "code-release",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/1904.01561",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "Chemprop v2.x",
     "id_or_null": "model-chemprop-v2",
     "first_release": {
      "date": "2024-04-23",
      "precision": "day",
      "year_hint": null,
      "basis": "GitHub v2.0.0 stable release",
      "event_kind": "code-release",
      "timestamp_utc": null,
      "source_url": "https://api.github.com/repos/chemprop/chemprop/releases/tags/v2.0.0",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "This record covers the v2.x rewrite only; the older version's benchmark results must not be attributed to v2"
  },
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "Open-source molecular property prediction library on directed message passing, the most common deep-learning baseline",
  "org": {
   "names": [
    "MIT (chemical engineering and electrical engineering groups)"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (graph neural network, directed message passing)",
  "open_status": {
   "code": "open",
   "weights": "closed",
   "weights_license": null,
   "code_license": "MIT",
   "license_changed": null,
   "note": "No general pretrained weights; users train on their own data, so any \"Chemprop score\" depends on who trained it and on what data"
  },
  "first_release": {
   "date": "2024-04-23",
   "precision": "day",
   "year_hint": null,
   "basis": "GitHub v2.0.0 stable release (first public software release)",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://api.github.com/repos/chemprop/chemprop/releases/tags/v2.0.0",
   "verification": "partial",
   "note": "The site treated paper dates (preprint 2025-09-15, journal 2025-12-26) as the release; a public pre-release v2.0.0-rc.1 (2024-03-04) is unverified"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-04-23",
    "precision": "day",
    "year_hint": null,
    "basis": "v2.0.0 stable release",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://api.github.com/repos/chemprop/chemprop/releases/tags/v2.0.0",
    "verification": "partial",
    "note": "Earlier pre-release and preprint evidence unverified"
   },
   "paper_v1": {
    "date": "2025-09-15",
    "precision": "day",
    "year_hint": null,
    "basis": "v2 paper preprint",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.cambridge.org/engage/coe/article-details/68bb5b5b23be8e43d65c5cc5",
    "verification": "partial",
    "note": null
   },
   "journal_online": {
    "date": "2025-12-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal version online",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1021/acs.jcim.5c02332",
    "verification": "partial",
    "note": null
   },
   "code_release": {
    "date": "2024-04-23",
    "precision": "day",
    "year_hint": null,
    "basis": "v2.0.0 stable release",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://api.github.com/repos/chemprop/chemprop/releases/tags/v2.0.0",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-08-04",
   "precision": "day",
   "year_hint": null,
   "basis": "GitHub v2.3.1 patch release",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://api.github.com/repos/chemprop/chemprop/releases",
   "verification": "verified",
   "description": "v2.3.1 patch, preceded by v2.3.0 (2026-07-08) and the v2.2.x series",
   "note": null
  },
  "status": "active",
  "status_note": "Several releases within 12 months",
  "what_it_does": "Treats a molecule as a graph, builds a molecular vector by message passing on directed edges and feeds a feed-forward head for regression or classification; v2 is a full rewrite with a Python API and multi-GPU training, and the paper claims algorithmic equivalence to the previous generation with higher speed and lower memory rather than new accuracy records. It is not a pretrained foundation model but a train-on-your-own-data tool, so it usually appears as a baseline: ADMET-AI builds on it and KERMT uses its multitask variant as a comparison. Next to pretrained models such as MolE, MiniMol and Uni-Mol2 it is small, fast to train and easy to retrain on in-house data.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "No fixed training set; the downstream ADMET build uses 41 ADMET datasets from TDC",
    "source_url": "https://github.com/chemprop/chemprop"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The library ships no data; users may train on in-house data, as corporate comparisons do",
    "source_url": "https://github.com/chemprop/chemprop"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-03-moleculenet",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://arxiv.org/abs/1904.01561",
    "confidence": "high",
    "second_hand": false,
    "note": "Those results belong to the 2019 original; the v2 abstract only says datasets and splits follow the previous software paper and that the algorithm is equivalent, without stating a re-run, so it is not treated as reported. Provenance is incomplete (missing url and date) and the entry is withheld from the ranking row"
   },
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://tdcommons.ai/benchmark/admet_group/01caco2/",
    "confidence": "low",
    "second_hand": true,
    "note": "Board entries carry no version label and appear indirectly through the ADMET model built on it; the library itself reports no scores"
   },
   {
    "benchmark_id": "bench-27-polaris",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "evidence_url": "https://chemrxiv.org/engage/chemrxiv/article-details/6841f19f",
    "confidence": "medium",
    "second_hand": false,
    "note": "Appears as an anonymous baseline in an organiser report with no per-item values"
   }
  ],
  "other_evaluations": [
   {
    "label": "Tasks in the v2 paper (reaction barriers, absorption wavelengths, a large quantum-property set)",
    "category": "public-benchmark-not-on-site",
    "detail": "The abstract claims speed and memory gains with accuracy unchanged from the previous generation, and gives no per-item values",
    "source_type": "self"
   },
   {
    "label": "Downstream applications such as antibiotic discovery",
    "category": "case-study",
    "detail": "Several studies use the library as a model component; applications rather than evaluations of the library",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: v2 is algorithmically equivalent to the previous generation while roughly halving runtime and cutting memory to about a third, with a Python API and multi-GPU training.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Third parties use it as a baseline: a competition organiser report lists it as an anonymous baseline and a public ADMET board carries entries built on it, but neither is an independent accuracy evaluation of v2.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "An academic open-source project from two engineering departments at a US technical university, not a company; the code lives under a same-named GitHub organisation under the MIT licence. The first version shipped with a 2019 methods paper, a software paper followed in 2023 and v2 is the rewrite.",
   "funding": "The v2 paper acknowledges a pharmaceutical machine-learning industry consortium, two US federal graduate fellowship programmes, a defence research programme and fellowships from a software company and a pharma company.",
   "products_partnerships": "No commercial entity; the ecosystem includes an ADMET prediction platform built on it, corporate research using its multitask variant as a comparison, and applications such as antibiotic discovery.",
   "related_models": [
    "model-admet-ai",
    "model-kermt",
    "model-mole",
    "model-minimol"
   ]
  },
  "reading_notes": [
   "The paper's claim is engineering - equivalent, faster, easier - not a new accuracy record.",
   "Old MoleculeNet numbers belong to the first version and must not be attributed to v2.",
   "With no general pretrained weights, a \"Chemprop score\" depends on who trained it and on what data, so cross-paper comparison requires reading the setup.",
   "Under this round's rules none of the three pairings count - two unconfirmed, one comparison-only - so the model enters the zero-pairing list."
  ],
  "caveats": [
   "Enters the zero-pairing list this round, recomputed under the new relation rules.",
   "The site treated paper dates as first release; the software v2.0.0 actually shipped 2024-04-23.",
   "An earlier public pre-release (2024-03-04) is unverified, so first release is flagged.",
   "Entries on the public ADMET board carry no version label, so they cannot be confirmed as v2."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1021/acs.jcim.5c02332",
    "title": "v2 journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.cambridge.org/engage/coe/article-details/68bb5b5b23be8e43d65c5cc5",
    "title": "v2 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/chemprop/chemprop",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://api.github.com/repos/chemprop/chemprop/releases",
    "title": "Release list",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/1904.01561",
    "title": "Original methods paper (2019)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "title": "Public ADMET leaderboard overview",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-09-15 preprint / 2025-12-26 journal",
    "now": "2024-04-23 (software v2.0.0 stable release); the paper date is listed separately as a milestone",
    "source_url": "https://api.github.com/repos/chemprop/chemprop/releases/tags/v2.0.0",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "TDC recorded as an official leaderboard",
    "now": "Unconfirmed: the board entry has no version label and appears only indirectly through a downstream model",
    "source_url": "https://tdcommons.ai/benchmark/admet_group/01caco2/",
    "date_found": "2026-11-16"
   },
   {
    "field": "n_benchmarks_reported",
    "was": "Counted on 3",
    "now": "Counted on 0; the model joins the zero-pairing list",
    "source_url": "https://chemrxiv.org/engage/chemrxiv/article-details/6841f19f",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.first_public_evidence",
   "benchmarks_reported[0]",
   "benchmarks_reported[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-chemprop-v2/",
   "en": "/bio/benchmark/models/chemprop-v2/"
  },
  "names_full_null_reason": "The software name has no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 2,
    "cited-as-comparison-only": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-claude-protein-design",
  "kind": "model",
  "short_slug": "claude-protein-design",
  "names": {
   "display": "Claude protein-design study (general model + public tools)",
   "full": "Study of a general language model acting as a design agent over public protein-design and co-folding tools",
   "aliases": [],
   "ambiguous_aliases": [
    "Claude"
   ]
  },
  "family": null,
  "family_null_reason": "The subject is a study configuration, not a model family; the same partners' separate 2025-10 study is listed on its own",
  "domain": "protein",
  "domains_secondary": [
   "text-agents",
   "generation"
  ],
  "one_liner": "A general language model orchestrates public tools to design binders, tested by outside labs",
  "org": {
   "names": [
    "Anthropic",
    "Adaptyv Bio",
    "Twist Bioscience"
   ],
   "global_entity_ids": [
    "adaptyv-bio"
   ]
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "US",
   "CH"
  ],
  "model_class": "Agent (general model orchestrating public design tools)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The general model is closed and no code was released; prompts, computational models of the designed complexes and all in vitro and computational data are public as a dataset"
  },
  "first_release": {
   "date": "2026-08-18",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer research post published (a partner case study followed the next day)",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://www.anthropic.com/research/Claude-accelerates-protein-design",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-08-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer research post",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://www.anthropic.com/research/Claude-accelerates-protein-design",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-08-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer research post, releasing prompts, computational models and in vitro data",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://www.anthropic.com/research/Claude-accelerates-protein-design",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-08-19",
   "precision": "day",
   "year_hint": null,
   "basis": "Partner laboratory case-study report published",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
   "verification": "verified",
   "description": "The latest sourced event is the partner laboratory's case-study report; a joint competition ran 2026-09-28 to 10-31 with results due 2026-12-15 and nothing citable yet",
   "note": null
  },
  "status": "product-only",
  "status_note": "No public model artefact: the general model is closed and what is public is prompts and datasets",
  "what_it_does": "Not a new generative network: a general language model acts as an agent, orchestrating public backbone-generation, sequence-design and co-folding filtering tools per target, after which two outside organisations express the designs and test binding. Unlike BindCraft or BoltzGen the contribution is orchestration and selection; the tools called include RFdiffusion2/3, BoltzGen, Proteina-Complexa, Protenix and ESMFold2, though the tool list appears only in second-hand summaries.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "The general model's training data is not public; each public tool it calls has its own training data",
    "source_url": "https://www.anthropic.com/research/Claude-accelerates-protein-design"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No dedicated training in this study; wet-lab results were not fed back",
    "source_url": "https://www.anthropic.com/research/Claude-accelerates-protein-design"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Two outside organisations handled expression and binding assays, for validation only",
    "source_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Partner-run external test (blinded submission)",
    "targets": 15,
    "designs_tested": 1320,
    "hits": 354,
    "denominator_basis": "1,440 designs on 16 targets, of which 15 targets and 1,320 designs have usable data (one target's 120 designs excluded for aggregation and uncertain data); expression 1,235 of 1,296 tested (95.3%)",
    "success_rate": "354/1,320 (final call pooled across two vendors; one vendor's raw calls give 336); 22.6%-35.1% by configuration",
    "affinity_best": null,
    "version": "dataset + partner report",
    "source_type": "third",
    "source_url": "https://huggingface.co/datasets/Anthropic/claude-protein-binder-design",
    "date": {
     "date": "2026-08-19",
     "precision": "day",
     "year_hint": null,
     "basis": "Partner case-study report",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Executed by a partner laboratory and a gene-synthesis vendor; the organiser is an interested party, so this is not an independent evaluation",
    "unverified_marker": null,
    "note": "Attributing 354 to a single vendor is inaccurate: it is the pooled two-vendor final call, and per-configuration rates must be listed separately rather than merged"
   },
   {
    "campaign": "Affinity comparison with past competition winners",
    "targets": 6,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Best design compared with the winning binder of each of six past competitions",
    "success_rate": "best designs showed greater affinity than five of the six past winners",
    "affinity_best": null,
    "version": "partner report",
    "source_type": "third",
    "source_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
    "date": {
     "date": "2026-08-19",
     "precision": "day",
     "year_hint": null,
     "basis": "Partner case-study report",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Executed and compiled by the partner laboratory",
    "unverified_marker": null,
    "note": "The published \"would have won five of six competitions\" overstates the source: it is a best-binder affinity comparison, not a ranking simulation, and one competition's best binder was not beaten"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-28-adaptyv",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
    "confidence": "high",
    "second_hand": false,
    "note": "The organiser ran a blinded external test on 16 targets from past competitions, hackathons and its benchmark set; not an entry in a public competition round"
   }
  ],
  "other_evaluations": [
   {
    "label": "Per-target and per-configuration hit rates",
    "category": "own-task-set",
    "detail": "Both parties report hit rates per target and per design configuration with wide variation; some per-target figures appear only in second-hand coverage",
    "source_type": "third"
   },
   {
    "label": "The same partners' separate 2025-10 binder study",
    "category": "case-study",
    "detail": "An earlier study applied the same approach to a different target class and sent designs for testing; it is a separate study and its figures are not merged here",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a general model can autonomously orchestrate public tools through the whole design pipeline, yielding binding designs for most targets at a high expression rate.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The wet lab was executed and compiled by an interested partner that also chose the targets; no disinterested third-party evaluation was found",
  "background": {
   "org_profile": "The design side is a US public-benefit AI company founded in 2021 and based in San Francisco (the work is credited to its science team, with no individuals listed); the wet-lab side is a European protein measurement service company and a US gene-synthesis company, with another vendor supplying a high-throughput surface plasmon resonance platform.",
   "funding": "The design side announced a $30B Series G in 2026-02 at a $380B post-money valuation, co-led by several investors (second-hand reporting).",
   "products_partnerships": "The products are the company's general model family and its science programme; with the wet-lab partner it also ran a joint competition 2026-09-28 to 10-31, with results due 2026-12-15.",
   "related_models": [
    "model-bindcraft",
    "model-boltzgen",
    "model-rfdiffusion2-3",
    "model-proteina-complexa"
   ]
  },
  "reading_notes": [
   "The comparison is against historical competition data rather than a matched control, and the interested organiser chose the targets.",
   "The pooled hit rate swings widely with target difficulty, including a zero-hit target, so it is not a stable rate.",
   "Only binding was measured; there are no functional, specificity or developability conclusions.",
   "The tool list exists only in second-hand summaries, so the gain cannot be split between the orchestration and the tools called."
  ],
  "caveats": [
   "The final hit call pools two vendors; one vendor's raw calls differ.",
   "The \"would have won five of six\" framing is wrong; correctly, the best designs out-bound five of the winners by affinity.",
   "The entry is retrospective dataset use and does not count toward reported benchmarks.",
   "The joint competition's results are not out and cannot be cited."
  ],
  "sources": [
   {
    "url": "https://www.anthropic.com/research/Claude-accelerates-protein-design",
    "title": "Developer research post",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
    "title": "Partner laboratory case study",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://huggingface.co/datasets/Anthropic/claude-protein-binder-design",
    "title": "Public dataset (prompts, computational models, in vitro data)",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://the-decoder.com/anthropic-says-any-lab-can-now-let-a-language-model-agent-run-the-whole-protein-design-stack/",
    "title": "Second-hand summary of the tool list and per-target figures",
    "kind": "second-hand",
    "date": null
   },
   {
    "url": "https://www.selectscience.net/article/carterra-enables-anthropic-s-autonomous-ai-protein-design-study-via-its-ht-spr-binding-platforms",
    "title": "Assay platform vendor coverage",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "354 binders measured by a single vendor; one hit rate of 26.8%",
    "now": "354/1,320 is the final call combining two vendors (336 for a single vendor); expression 1,235/1,296 = 95.3%; per-configuration 22.6%–35.1% listed separately",
    "source_url": "https://huggingface.co/datasets/Anthropic/claude-protein-binder-design",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Could have won 5 of 6 competitions",
    "now": "The best design's affinity beats 5 of the 6 winners, and does not exceed the best binder of one of them",
    "source_url": "https://adaptyvbio.substack.com/p/case-study-benchmarking-claudes-protein",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "training_data.a_public",
   "training_data.b_private_licensed",
   "background.funding",
   "other_evaluations[0]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-claude-protein-design/",
   "en": "/bio/benchmark/models/claude-protein-design/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has no page for this study (the partner organisation has its own entity)",
  "status_head": null
 },
 {
  "id": "model-co-scientist",
  "kind": "model",
  "short_slug": "co-scientist",
  "names": {
   "display": "Co-Scientist",
   "full": "AI co-scientist",
   "aliases": [
    "AI co-scientist",
    "Google Co-Scientist"
   ],
   "ambiguous_aliases": [
    "Co-Scientist",
    "co-scientist"
   ]
  },
  "family": null,
  "family_null_reason": "A single system; the 2025 preprint and 2026 journal version are versions within one record",
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "Google DeepMind's Gemini-based multi-agent system that generates and ranks research hypotheses",
  "org": {
   "names": [
    "Google DeepMind",
    "Google Research"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB",
   "US"
  ],
  "model_class": "Agent (hypothesis generation)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Closed; individual researchers apply via Gemini for Science and an enterprise version is in Google Cloud preview"
  },
  "first_release": {
   "date": "2025-02-26",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv preprint v1 (official blog the same period)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2502.18864",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-02-26",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2502.18864",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-02-26",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2502.18864",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-05-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Nature publication (official blog the same day)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://www.nature.com/articles/s41586-026-10644-y",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-02",
    "precision": "month",
    "year_hint": null,
    "basis": "Google Research blog",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-06-29",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv v2",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://export.arxiv.org/api/query?id_list=2502.18864",
   "verification": "verified",
   "description": "arXiv v2 posted (scope of revision not inspected), after the 2026-05-19 Nature paper and opening to individual researchers through Gemini for Science",
   "note": null
  },
  "status": "product-only",
  "status_note": "Closed; offered through an experimental tool and a cloud preview",
  "what_it_does": "A Gemini-based multi-agent system in which generation, reflection, ranking, evolution and meta-review agents iterate on hypotheses, ranking them in an Elo tournament of pairwise comparisons and simulated debates, and output research proposals; checks call web search and public databases such as ChEMBL and UniProt. Unlike Kosmos or Biomni, it outputs ranked hypotheses rather than running data analysis.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "Base Gemini training data undisclosed; ChEMBL, UniProt and similar are inference-time tools",
    "source_url": "https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "The system is not trained on wet-lab data",
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Some hypotheses were tested in partner labs; the developers give no aggregate counts",
    "source_url": "https://arxiv.org/abs/2502.18864"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Partner-lab case studies",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "No aggregate counts; individual cases such as liver-fibrosis targets, AML drug repurposing and cellular senescence, each run by a partner lab",
    "success_rate": null,
    "affinity_best": null,
    "version": "paper/blog",
    "source_type": "self",
    "source_url": "https://arxiv.org/abs/2502.18864",
    "date": {
     "date": "2025-02-26",
     "precision": "day",
     "year_hint": null,
     "basis": "Preprint v1",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2502.18864",
     "verification": "verified",
     "note": null
    },
    "verification": "unverified",
    "executor": "Run by external partner labs; not independent replication",
    "unverified_marker": "unverified",
    "note": "Qualitative cases only; must not be converted into hit rates or quantitative validation"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Site text and agent benchmarks",
    "category": "none-disclosed",
    "detail": "Checked LAB-Bench, BixBench, BioASQ, PubMedQA, BioML-bench, BioMysteryBench: not reported in the paper or official blog",
    "source_type": "self"
   },
   {
    "label": "Elo vs GPQA correctness",
    "category": "public-benchmark-not-on-site",
    "detail": "Uses GPQA Diamond to check whether system Elo tracks answer correctness; the Elo is self-assessed with no independent ground truth",
    "source_type": "self"
   },
   {
    "label": "Expert evaluation",
    "category": "user-or-partner-report",
    "detail": "A small expert panel rated novelty and impact; the developers describe the sample as small",
    "source_type": "self"
   },
   {
    "label": "Elo tournament",
    "category": "own-task-set",
    "detail": "Self-play ranking improves with test-time compute; no independent ground truth",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a multi-agent Gemini system that generates and ranks hypotheses, with Elo tracking quality and some hypotheses confirmed in partner labs.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Press reports describe a UK university team trying it on a bacterial-evolution question, and 2025 press criticism noted a lack of independent evaluation; no systematic third-party study was found.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Google DeepMind was founded in London in 2010, acquired by Google in 2014 and merged with Google Brain in 2023-04; this system was built with Google Research, Google Cloud and Google Labs.",
   "funding": "A wholly owned internal research unit of Alphabet with no separate funding.",
   "products_partnerships": "Enterprise preview partners include Daiichi Sankyo, Bayer Crop Science and US national laboratories; academic cases include Stanford, MIT, Edinburgh, Cambridge and Calico; the developers cite researchers at over 100 institutions.",
   "related_models": [
    "model-kosmos",
    "model-biomni",
    "model-robin",
    "model-medgemma-med-gemini",
    "model-verid"
   ]
  },
  "reading_notes": [
   "It has no standard site benchmark score and cannot be compared directly with Kosmos or Biomni.",
   "Partner-lab results are single cases without stated controls or replicates and cannot be read as efficacy or hit rates.",
   "The Elo-GPQA correlation is self-assessed and comes from the 2025 preprint.",
   "The 2026 claim of planning experiments has no detail in official text, so it is best read as a hypothesis-generation system."
  ],
  "caveats": [
   "Wet lab was done by partner labs with no aggregate counts; treated as unverified.",
   "Closed, so external evaluation is not possible.",
   "The scope of the arXiv v2 revision was not inspected."
  ],
  "sources": [
   {
    "url": "https://www.nature.com/articles/s41586-026-10644-y",
    "title": "Nature paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2502.18864",
    "title": "arXiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://deepmind.google/blog/co-scientist-a-multi-agent-ai-partner-to-accelerate-research/",
    "title": "DeepMind official blog (2026-05-19)",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/",
    "title": "Google Research 2025 blog",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://blog.google/innovation-and-ai/technology/research/gemini-for-science-io-2026",
    "title": "Gemini for Science launch blog",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1038/s41586-026-10644-y",
    "title": "Nature paper registry record",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-05-19 (Nature publication)",
    "now": "2026-06-29（arXiv v2）",
    "source_url": "https://export.arxiv.org/api/query?id_list=2502.18864",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "journal-publication",
    "now": "paper-revision",
    "source_url": "https://export.arxiv.org/api/query?id_list=2502.18864",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[0]",
   "release_milestones.official_announcement",
   "training_data.a_public"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-co-scientist/",
   "en": "/bio/benchmark/models/co-scientist/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-diffdock",
  "kind": "model",
  "short_slug": "diffdock",
  "names": {
   "display": "DiffDock",
   "full": null,
   "full_null_reason": "Official materials use DiffDock only; no expanded name",
   "aliases": [
    "DiffDock",
    "DiffDock-L"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "The record covers DiffDock and its in-repo successor DiffDock-L without separate family members",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Diffusion-based blind docking, now used mainly as the deep-learning docking baseline",
  "org": {
   "names": [
    "MIT CSAIL"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Blind docking (diffusion plus confidence ranking)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code and weights are public on GitHub, with DiffDock-L as the repository default"
  },
  "first_release": {
   "date": "2022-10-04",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv v1 (conference version the following year)",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2210.01776",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2022-10-04",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2210.01776",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2022-10-04",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2210.01776",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-02-28",
    "precision": "day",
    "event_kind": "new-version",
    "basis": "GitHub v1.1 (DiffDock-L)",
    "timestamp_utc": null,
    "source_url": "https://github.com/gcorso/DiffDock/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2024-02-28",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "DiffDock-L weights shipped with v1.1",
    "timestamp_utc": null,
    "source_url": "https://github.com/gcorso/DiffDock/releases",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2024-02-28",
   "precision": "day",
   "event_kind": "new-version",
   "basis": "GitHub v1.1: DiffDock-L release",
   "description": "DiffDock-L was trained with larger augmented data and confidence bootstrapping and appeared at that year's conference; v1.1.1-v1.1.3 (to 2024-09-04) are fixes and the last commit is 2025-05-02",
   "timestamp_utc": null,
   "source_url": "https://github.com/gcorso/DiffDock/releases",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "No newer generation; later activity is maintenance",
  "what_it_does": "Treats docking as generation: it diffuses over ligand translation, rotation and torsion, then ranks candidate poses with a confidence model, and DiffDock-L is the larger-data successor. In the co-folding era it serves as the deep-learning docking baseline against the physics-scoring line of Vina and against AlphaFold3 and NeuralPLexer3.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDBBind, with DiffDock-L adding other PDB complexes and synthetic augmentation that cuts ligands from real structures",
    "source_url": "https://arxiv.org/abs/2402.18396"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "PoseBusters V2 comparison table in a third-party paper",
    "source_url": "https://arxiv.org/abs/2412.10743",
    "confidence": "medium",
    "verification": "verified",
    "note": "The number comes from another paper's comparison table, not from the developers, and is not the successor's official result"
   },
   {
    "benchmark_id": "bench-12-dude",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "LIT-PCBA virtual-screening study by a third party",
    "source_url": "https://arxiv.org/abs/2605.01681",
    "confidence": "medium",
    "verification": "verified",
    "note": "A third-party study ran this family on a screening task; strong pose generation does not imply strong screening"
   },
   {
    "benchmark_id": "bench-33-posex",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "PoseX, rerun by the benchmark authors",
    "source_url": "https://arxiv.org/html/2505.01700v3",
    "confidence": "medium",
    "verification": "verified",
    "note": "A pairing added this round, evaluated by the PoseX paper itself"
   }
  ],
  "other_evaluations": [
   {
    "label": "Self-reported blind-docking sets",
    "category": "self-report",
    "detail": "The DiffDock-L paper self-reports top-1 results on two blind-docking sets under a fixed sampling budget, a different protocol from physical-validity checks (no figures)",
    "source_type": "self"
   },
   {
    "label": "Second-hand reviews",
    "category": "second-hand",
    "detail": "The claim that it beats other blind-docking methods partly reaches readers through reviews rather than independent reruns (second-hand)",
    "source_type": "third"
   },
   {
    "label": "Baseline role",
    "category": "third-party-evaluation",
    "detail": "Several co-folding and docking papers rerun it as the deep-learning baseline, with conclusions that vary by task",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report that DiffDock-L, trained with larger augmented data and confidence bootstrapping, beats other blind-docking methods (self-reported, partly relayed by reviews).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Third-party papers and benchmark authors have rerun it; the conclusion is a baseline role, with no edge in physical validity or screening (no figures).",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "An academic group with no company: teams at MIT CSAIL and an affiliated clinic, whose authors were then mostly doctoral students and faculty at the institute.",
   "funding": null,
   "funding_null_reason": "Academic project; funding amounts undisclosed",
   "products_partnerships": "The successor DiffDock-L is the repository default; it shares an academic lineage with Boltz-2 but no institutional link is evident.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-vina",
    "model-boltz-2",
    "model-neuralplexer3",
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "Most of its benchmark figures come from other papers under mixed protocols and should be read as a baseline only.",
   "Its self-reported blind-docking protocol differs from physical-validity checks; do not swap them.",
   "It has no edge in virtual screening; strong pose generation does not imply strong screening.",
   "Classical docking scores higher when the pocket is known, but the settings differ."
  ],
  "caveats": [
   "The developers self-report on none of the site's benchmarks; all three relations come from third parties.",
   "The PoseBusters line is cited as comparison only and does not count as an established relation.",
   "The last substantive update was 2024-02, followed only by fixes and maintenance."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2210.01776",
    "title": "DiffDock paper",
    "kind": "preprint",
    "date": {
     "date": "2022-10-04",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "arXiv v1",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2210.01776",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://arxiv.org/abs/2402.18396",
    "title": "DiffDock-L paper",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/gcorso/DiffDock/releases",
    "title": "DiffDock GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2412.10743",
    "title": "Third-party comparison table in a co-folding paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2605.01681",
    "title": "Third-party virtual-screening study",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2505.01700v3",
    "title": "Evaluation paper by the benchmark authors",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-diffdock/",
   "en": "/bio/benchmark/models/diffdock/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-dplm-2",
  "kind": "model",
  "short_slug": "dplm-2",
  "names": {
   "display": "DPLM-2",
   "full": null,
   "full_null_reason": "The paper uses the model name only, with no expanded form",
   "aliases": [
    "DPLM-2"
   ],
   "ambiguous_aliases": [
    "DPLM",
    "DPLM-2.1"
   ]
  },
  "family": null,
  "family_null_reason": "The sequence-only predecessor and DPLM-2.1 are separate items; this record covers DPLM-2 only",
  "domain": "protein",
  "domains_secondary": [
   "generation",
   "structure"
  ],
  "one_liner": "Multimodal discrete-diffusion protein language model generating matched sequence and structure",
  "org": {
   "names": [
    "ByteDance Research",
    "Nanjing University"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": "A joint project of an internet company's research arm and a university computer-science department, with no separate corporate entity",
  "country": [
   "CN"
  ],
  "model_class": "Design (joint sequence-structure diffusion)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": "Apache-2.0 (per a mirror site, unverified)",
   "license_changed": null,
   "note": "Code and weights are public and the repository is archived read-only; the licence label comes from a mirror site (unverified)"
  },
  "first_release": {
   "date": "2024-10-17",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "Preprint v1 submission",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2410.13782",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-10-17",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2410.13782",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-10-17",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2410.13782",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-05-15",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Repository commit \"release of dplm-2\"",
    "timestamp_utc": null,
    "source_url": "https://github.com/bytedance/dplm",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-10-17",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Project page published with the preprint",
    "timestamp_utc": null,
    "source_url": "https://bytedance.github.io/dplm",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2025-07-25",
   "precision": "day",
   "event_kind": "code-maintenance-only",
   "basis": "The final commit before the repository was archived",
   "description": "The final commit tweaked a sampling default, after which the repository became read-only (the archive date itself is unsourced); the successor DPLM-2.1 paper is listed separately and is not an update to this model",
   "timestamp_utc": null,
   "source_url": "https://github.com/bytedance/dplm",
   "verification": "partial",
   "note": "No source found for the archive date (unverified)"
  },
  "status": "archived",
  "status_note": "Repository archived read-only; development moved to the successor",
  "what_it_does": "Turns 3D coordinates into discrete structure tokens with lookup-free quantization and runs discrete diffusion over the joint sequence-structure distribution, generating matched sequence and structure in one pass while supporting folding, inverse folding and motif scaffolding; it is warm-started from the sequence-only predecessor. ESM3 is the comparable joint sequence-structure model on this site.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public sequence and experimental-structure databases, an annotated protein database and synthetic structures",
    "source_url": "https://arxiv.org/abs/2410.13782"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data reported",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-15-cameo",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "frozen 2022 target set, folding and inverse folding",
    "source_url": "https://arxiv.org/html/2410.13782v1",
    "confidence": "high",
    "verification": "partial",
    "note": "Retrospective dataset use: a frozen snapshot of a released target set from the service, with no live submission and no service-assigned ranking, so it must not be read as participation"
   }
  ],
  "other_evaluations": [
   {
    "label": "Structure prediction (folding)",
    "category": "public-benchmark-not-on-site",
    "detail": "Sequence-to-structure on a frozen target set and a date-split structure database, against several folding and multimodal models",
    "source_type": "self"
   },
   {
    "label": "Inverse folding",
    "category": "public-benchmark-not-on-site",
    "detail": "Amino-acid recovery and self-consistency on the same frozen target set and date split",
    "source_type": "self"
   },
   {
    "label": "Unconditional co-generation",
    "category": "comparison-vs-baseline-only",
    "detail": "One hundred samples at each of five lengths, comparing designability and diversity against several generative models",
    "source_type": "self"
   },
   {
    "label": "Motif scaffolding",
    "category": "public-benchmark-not-on-site",
    "detail": "Twenty-four motif problems taken from prior work with 100 scaffolds each, against several generative models",
    "source_type": "self"
   },
   {
    "label": "Representation-learning tasks",
    "category": "public-benchmark-not-on-site",
    "detail": "Localization, thermostability, protein interaction and metal-ion binding against the predecessor and others, not winning on all of them",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that one multimodal sequence-structure diffusion model handles folding, inverse folding and motif scaffolding competitively, with diverse and structurally plausible co-generated samples, and that structure-aware representations also help predictive tasks.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation or independent replication found",
  "background": {
   "org_profile": "A big-tech lab and university joint project with no separate corporate entity. The research arm used one name on the 2024 paper and a different in-group research brand on the 2025 successor paper (same team); the partner is a university computer-science department.",
   "funding": null,
   "funding_null_reason": "Funded from the company's internal R&D budget; no external grants itemised",
   "products_partnerships": "The predecessor is the sequence-only discrete-diffusion protein language model; the successor DPLM-2.1 (preprint v1 2025-04, v3 2025-06) explores the multimodal design space and is recorded separately from this model's repository events.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-esm3",
    "model-la-proteina",
    "model-rfdiffusion2-3"
   ]
  },
  "reading_notes": [
   "Its only site pairing is retrospective dataset use on a frozen target set: not participation, and with no service-assigned ranking.",
   "All evaluation is computational, with no wet-lab validation.",
   "The value is methodological: it shows discrete diffusion can unify sequence and structure generation.",
   "Synthetic structures enter training, so generation quality is capped by the model that produced them."
  ],
  "caveats": [
   "The repository is archived read-only, so expect no further fixes; the successor has its own entry.",
   "The archive date is unsourced (unverified), so the latest update is the final commit before archiving.",
   "Under this round's rules it leaves the zero-pairing list, but that pairing is retrospective reuse and must not be written as participation.",
   "The licence label comes from a mirror site (unverified)."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2410.13782",
    "title": "DPLM-2 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2410.13782v1",
    "title": "Preprint v1 full text with the evaluation setup",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/bytedance/dplm",
    "title": "Official repository (archived)",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://bytedance.github.io/dplm",
    "title": "Project page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2504.11454",
    "title": "Successor DPLM-2.1 preprint",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.date",
    "was": "2025-04-15, the date of the successor paper DPLM-2.1 v1",
    "now": "2025-07-25, the final commit before this model's repository was archived, with the successor paper recorded separately",
    "source_url": "https://github.com/bytedance/dplm",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "Recorded as a successor model",
    "now": "Changed to code maintenance only: the event is a repository commit, not a new model",
    "source_url": "https://github.com/bytedance/dplm",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "The site said it reported no site benchmark and listed it as zero-pairing",
    "now": "Under this round's rules it has a retrospective-dataset-use relation with one evaluation service and therefore leaves the zero-pairing list",
    "source_url": "https://arxiv.org/html/2410.13782v1",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "open_status.code_license",
   "release_milestones.code_release",
   "release_milestones.official_announcement",
   "benchmarks_reported[0]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-dplm-2/",
   "en": "/bio/benchmark/models/dplm-2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-dyno-psi",
  "kind": "model",
  "short_slug": "dyno-psi",
  "names": {
   "display": "Dyno Psi-1 / Psi-Phi",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "Dyno Psi",
    "DynoPsi",
    "Psi-1",
    "Psi-Phi"
   ],
   "ambiguous_aliases": []
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Psi-1",
     "id_or_null": null,
     "first_release": {
      "date": "2026-03-18",
      "precision": "day",
      "event_kind": "code-release",
      "basis": "GitHub v1.0.0 and the whitepaper on the same day",
      "timestamp_utc": null,
      "source_url": "https://github.com/dynotx/dynopsi/releases/tag/v1.0.0",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "Psi-Phi",
     "id_or_null": null,
     "first_release": {
      "date": "2026-03-18",
      "precision": "day",
      "event_kind": "code-release",
      "basis": "Companion filter and agent suite released with Psi-1",
      "timestamp_utc": null,
      "source_url": "https://github.com/dynotx/dynopsi/releases/tag/v1.0.0",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "Psi-1.5 Cicada",
     "id_or_null": null,
     "first_release": {
      "date": "2026-11-13",
      "precision": "day",
      "event_kind": "new-weights",
      "basis": "Version 2.0.0 switched the default weights to this model",
      "timestamp_utc": null,
      "source_url": "https://github.com/dynotx/dynopsi/releases/tag/v2.0.0",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "Covers the Psi generator series and the companion filter/agent suite Psi-Phi; the wet-lab arm used the earlier Psi-0.5, which is separate from Psi-1"
  },
  "family_null_reason": null,
  "domain": "protein",
  "domains_secondary": [],
  "one_liner": "Open-weight binder design model of about 200M parameters with a companion filter and agent suite",
  "org": {
   "names": [
    "Dyno Therapeutics"
   ],
   "global_entity_ids": []
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Binder backbone generation",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Weights and code Apache-2.0; the synthetic training dataset CC BY 4.0"
  },
  "first_release": {
   "date": "2026-03-18",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub v1.0.0 with the whitepaper and press release on the same day (conference launch)",
   "timestamp_utc": null,
   "source_url": "https://github.com/dynotx/dynopsi/releases/tag/v1.0.0",
   "verification": "verified",
   "note": "Repository creation dates are not used as first release; the press-release date was not independently re-checked"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2026-03-18",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub v1.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/dynotx/dynopsi/releases/tag/v1.0.0",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-03-18",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights released with v1.0.0",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/api/models/dynotx/dynopsi",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-11-13",
   "precision": "day",
   "event_kind": "new-version",
   "basis": "GitHub release 2.0.0",
   "description": "Version 2.0.0: default weights switched to Psi-1.5 Cicada, with breaking API changes",
   "timestamp_utc": null,
   "source_url": "https://github.com/dynotx/dynopsi/releases/tag/v2.0.0",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "New version and new weights within 12 months",
  "what_it_does": "An open-weight binder design model: a flow-matching backbone generator of about 200M parameters followed by a sequence-design tool, with Psi-Phi as the companion filter and agent suite.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Experimental and predicted structure databases plus ~4M synthetic domain-interface dimers (32,253 high-confidence ones used for fine-tuning)",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "The extent of experimental feedback into training is not disclosed",
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Yeast display and SPR used for validation only",
    "source_url": null
   }
  },
  "wetlab_validation": [
   {
    "campaign": "IL-7RA yeast-display comparison",
    "modality": "miniprotein",
    "targets": null,
    "designs_tested": 100,
    "hits": null,
    "denominator_basis": "Initial sampling budget matched (~12k), not design counts: the comparator's full pipeline yielded 125 designs from that budget while these designs were subsampled from the same budget; success was measured on the top 100 designs ranked by ipSAE, with no per-target numbers, so no rate is printed",
    "success_rate": null,
    "affinity_best": null,
    "version": "Psi-0.5",
    "source_type": "self",
    "source_url": "https://dynopsi.dynotx.com/dynopsi_whitepaper.pdf",
    "date": null,
    "verification": "verified",
    "note": "The wet-lab arm is the earlier Psi-0.5, not Psi-1; developer-run, not an independent replication"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Five-target in-silico filter pass rate",
    "category": "own-task-set",
    "detail": "Psi-1 computational filtering: 16,900 designs per model across 13 length bins; best on only 2 of 5 targets, i.e. parity not a win; this is not a wet-lab result",
    "source_type": "self"
   },
   {
    "label": "Press-release comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "The company states an advantage over several open methods without per-target numbers",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that this ~200M open-weight model, with test-time scaling against high-throughput experiments, gives a higher IL-7RA wet-lab success rate than the comparator",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent head-to-head found",
  "background": {
   "org_profile": "Dyno Therapeutics is an AI-driven AAV capsid design company founded in Massachusetts in 2018",
   "funding": "Announced a $100M Series A in 2021-05 led by a16z",
   "funding_null_reason": null,
   "products_partnerships": "Capsid deals with large pharma companies (one licence exercised in 2026-04); the whitepaper was written by the company's AI lab with engineering support from a GPU company",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-la-proteina",
    "model-boltzgen",
    "model-rfdiffusion2-3"
   ]
  },
  "reading_notes": [
   "The five-target result is Psi-1 computational filtering, not a five-target wet-lab campaign",
   "The wet-lab comparison matches sampling budget rather than design counts, and used Psi-0.5",
   "Success was measured only on the top 100 ranked designs, with no per-target numbers"
  ],
  "caveats": [
   "No results reported on any of the 44 benchmarks",
   "The press release's advantage claim has no per-target data behind it"
  ],
  "sources": [
   {
    "url": "https://github.com/dynotx/dynopsi/releases/tag/v2.0.0",
    "title": "GitHub repository dynotx/dynopsi",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/dynotx/dynopsi/releases/tag/v1.0.0",
    "title": "GitHub repository dynotx/dynopsi",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/api/models/dynotx/dynopsi",
    "title": "Hugging Face page api/models",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://dynopsi.dynotx.com/dynopsi_whitepaper.pdf",
    "title": "Official page dynopsi.dynotx.com",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://huggingface.co/dynotx/dynopsi",
    "title": "Hugging Face page dynotx/dynopsi",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "Comparison at equal design counts",
    "now": "Initial sampling budget matched (~12k); the wet-lab arm is Psi-0.5",
    "source_url": "https://dynopsi.dynotx.com/dynopsi_whitepaper.pdf",
    "date_found": "2026-11-16"
   },
   {
    "field": "other_evaluations",
    "was": "Five-target wet-lab pass rate",
    "now": "Psi-1 computational filtering, best on only 2 of 5 targets",
    "source_url": "https://dynopsi.dynotx.com/dynopsi_whitepaper.pdf",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-dyno-psi/",
   "en": "/bio/benchmark/models/dyno-psi/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-enchant-v1-v2",
  "kind": "model",
  "short_slug": "enchant-v1-v2",
  "names": {
   "display": "Enchant v1/v2",
   "full": null,
   "aliases": [
    "Enchant v1",
    "Enchant v2",
    "Enchant v3"
   ],
   "ambiguous_aliases": [
    "Enchant"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Enchant v1",
     "id_or_null": null,
     "first_release": {
      "date": "2024-10-29",
      "precision": "day",
      "year_hint": null,
      "basis": "Company press release",
      "event_kind": "product-update",
      "timestamp_utc": null,
      "source_url": "https://www.iambic.ai/post/enchant",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "Enchant v2",
     "id_or_null": null,
     "first_release": {
      "date": "2025-05-05",
      "precision": "day",
      "year_hint": null,
      "basis": "Date stated on the official blog",
      "event_kind": "new-version",
      "timestamp_utc": null,
      "source_url": "https://iambic.ai/post/enchantv2",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "Enchant v3",
     "id_or_null": null,
     "first_release": {
      "date": "2026-09-21",
      "precision": "day",
      "year_hint": null,
      "basis": "Date on the official page",
      "event_kind": "new-version",
      "timestamp_utc": null,
      "source_url": "https://www.iambic.ai/news/enchant-3",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "Covers three generations of one closed product (v1/v2/v3), none with weights or third-party review"
  },
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [
   "text-agents",
   "structure"
  ],
  "one_liner": "Closed multimodal transformer predicting molecular properties and clinically relevant endpoints from early data",
  "org": {
   "names": [
    "Iambic Therapeutics"
   ],
   "global_entity_ids": [
    "iambic-therapeutics"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction / multimodal foundation model",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "All three generations are closed, disclosed only through company pages and press releases, with no weights and no cited API documentation"
  },
  "first_release": {
   "date": "2024-10-29",
   "precision": "day",
   "year_hint": null,
   "basis": "Company press release for v1 (earliest sourced public evidence)",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.businesswire.com/news/home/20241029837875/en/Iambic-Therapeutics-Announces-Enchant-an-AI-Platform-that-Predicts-Clinical-Outcomes-from-the-Earliest-Stages-of-Drug-Discovery",
   "verification": "verified",
   "note": "Product-only: no paper and no weights"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-10-29",
    "precision": "day",
    "year_hint": null,
    "basis": "v1 press release",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.iambic.ai/post/enchant",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-10-29",
    "precision": "day",
    "year_hint": null,
    "basis": "Company press release and official blog",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.iambic.ai/post/enchant",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09-21",
   "precision": "day",
   "year_hint": null,
   "basis": "v3 launch (official page)",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://www.iambic.ai/news/enchant-3",
   "verification": "verified",
   "description": "v3 launched: the company describes a 41B-parameter mixture-of-experts multimodal transformer (v1 1B, v2 7B) taking structures, sequences and omics, tabular assays, physics data, text and images, adding inverse design and peptide capabilities; the parameter framing matches the offering document. The developer's 2026-10-16 listing is a corporate event, not a model update",
   "note": "The 41B and mixture-of-experts details were not independently re-verified"
  },
  "status": "product-only",
  "status_note": "Closed, with no public weights or code; product-only",
  "what_it_does": "Puts molecular structures and several experimental and clinically relevant modalities into one transformer, aiming to close the preclinical-to-clinical data gap and predict ADME, pharmacokinetics and clinically relevant endpoints; v2 claims more than a tenfold scale-up over v1 and v3 adds inverse design, omics and image prediction. The company says it fine-tunes weekly across hundreds of molecular properties. TxGemma has been used as a comparator; TerraBind, backed by the same chip investor, instead trains on public data with its own wet-lab validation.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Said to come from dozens of external sources, none named; a public human pharmacokinetics dataset is used for the PK task",
    "source_url": "https://www.iambic.ai/post/enchant"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "Possibly present but undisclosed; the published site had a claim without scale",
    "source_url": "https://www.iambic.ai/post/enchant"
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "Says an in-house automated synthesis and assay platform generates new data continuously, with weekly fine-tuning across dozens of endpoints; scale undisclosed",
    "source_url": "https://iambic.ai/post/enchantv2"
   },
   "d_wetlab_validation_only": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://www.iambic.ai/post/enchant",
    "confidence": "high",
    "second_hand": false,
    "note": "The developer post never names the benchmark, the official split or the protocol - only a hyperlink to a baseline page. The zh matrix, zh model page and en data disagreed three ways; all are unified to unconfirmed and it is not counted"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://www.iambic.ai/news/enchant-3",
    "confidence": "low",
    "second_hand": false,
    "note": "The v3 page claims evaluation on held-out structures from that event, figure only and no protocol"
   },
   {
    "benchmark_id": "bench-15-cameo",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://www.iambic.ai/news/enchant-3",
    "confidence": "low",
    "second_hand": false,
    "note": "Same source, using the frozen 2022 static set with figures only; even if it held it would fall under retrospective dataset use, not participation"
   }
  ],
  "other_evaluations": [
   {
    "label": "Public human pharmacokinetics dataset (half-life)",
    "category": "public-benchmark-not-on-site",
    "detail": "The company reports better correlation than the prior best on the full dataset; the announcement links a baseline page without naming a benchmark and does not state the split",
    "source_type": "self"
   },
   {
    "label": "Low-data setting",
    "category": "comparison-vs-baseline-only",
    "detail": "v2 claims that training on 5% of the dataset (about two dozen molecules) beats the prior best trained on all of it",
    "source_type": "self"
   },
   {
    "label": "Company-internal task sets",
    "category": "own-task-set",
    "detail": "v2 compares against v1 and TxGemma on a kinase set, one company's ADME set, a cell pharmacodynamics set and rat in-vivo clearance, mostly as figures",
    "source_type": "self"
   },
   {
    "label": "v3 protein structure quality comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "The official page claims better quality than two public protein models on held-out structures, with figures only and no numbers",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a multimodal transformer pretrained on abundant preclinical data predicts human pharmacokinetics with very little clinical data and beats the prior best in low-data settings.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The model is closed with no weights and every figure is company-reported; no third-party evaluation or replication was found",
  "background": {
   "org_profile": "An AI small-molecule drug discovery company incorporated in the US in 2019 and renamed in 2023-05, headquartered in California with offices in the UK and Ireland; its offering document reports 171 full-time staff as of 2026-09 (140 R&D, 31 administrative). It runs its own pipeline plus a platform.",
   "funding": "A $100M Series B in 2023-10 (co-led by two funds, with a chip company and a sequencing company's venture arm as new investors); a $50M extension in 2024-06 (led by sovereign and family capital) taking the round above $150M; a further round of over $100M in 2025-11 with no lead named. The offering document reports about $461.8M raised since inception, $536.5M in total cash and $66.5M of convertible notes converting at listing. The IPO closed 2026-10-16 at $16.00 per share for 9,375,000 shares and $150M.",
   "products_partnerships": "The pipeline includes an oral brain-penetrant targeted inhibitor (in a phase 1/1b basket trial as of 2026-09), a programme that filed a clinical trial application on 2026-09-30 and one pre-filing programme; the platform comprises this model and a co-folding model. A 2026-09 collaboration and option-to-license deal with a large pharma company covers two initial targets with an upfront, milestones and tiered royalties, terms undisclosed; a 2026-02 research collaboration and licence with another pharma company covers three initial targets with aggregate milestones above $1.7B and $6.2M of revenue recognised in the first half of 2026; five further pharma and compute partners are named.",
   "related_models": [
    "model-txgemma",
    "model-terrabind",
    "model-neuralplexer3"
   ]
  },
  "reading_notes": [
   "Every figure is company-reported and the model is closed, so nothing can be independently reproduced.",
   "The TDC relation is ruled unconfirmed this round: the blog names no benchmark and no split, and all three published surfaces are unified to unconfirmed.",
   "The human pharmacokinetics correlation first appeared in the v1 announcement; the v2 blog restates the same figure without saying it was re-run, so it is not a v2 result.",
   "The real differentiator is continuously generated in-house wet-lab data in training, but its scale has never been disclosed, and v3's parameters and capabilities rest on company statements only."
  ],
  "caveats": [
   "No counted pairings: all three site relations are unconfirmed.",
   "Closed, with no weights or code and no third-party review.",
   "The scale of in-house wet-lab training data is undisclosed.",
   "Pretraining token and modality counts unique to the press release are not repeated on the official technical page or in the offering document."
  ],
  "sources": [
   {
    "url": "https://www.iambic.ai/post/enchant",
    "title": "Official v1 announcement",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://iambic.ai/post/enchantv2",
    "title": "Official v2 blog",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.iambic.ai/news/enchant-3",
    "title": "Official v3 page",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20241029837875/en/Iambic-Therapeutics-Announces-Enchant-an-AI-Platform-that-Predicts-Clinical-Outcomes-from-the-Earliest-Stages-of-Drug-Discovery",
    "title": "v1 press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.sec.gov/Archives/edgar/data/1997038/000119312526422865/iambic_inception_424b4.htm",
    "title": "Offering document (424B4)",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.sec.gov/Archives/edgar/data/1997038/000119312526424242/iam-20261016.htm",
    "title": "Current report on completion of the listing",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "TDC self-reported (Chinese matrix and English data)",
    "now": "Unified to unconfirmed and not counted: the blog does not name the benchmark or give the split and protocol",
    "source_url": "https://www.iambic.ai/post/enchant",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "v2 (2025-05-05) as the latest event",
    "now": "2026-09-21 v3 release; the company listing event is not counted as a model update",
    "source_url": "https://www.iambic.ai/news/enchant-3",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[0]",
   "benchmarks_reported[1]",
   "benchmarks_reported[2]",
   "latest_update.description",
   "training_data.b_private_licensed",
   "training_data.c_own_wetlab_for_training"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-enchant-v1-v2/",
   "en": "/bio/benchmark/models/enchant-v1-v2/"
  },
  "names_full_null_reason": "The product name has no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 3
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developing company but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-esm-c",
  "kind": "model",
  "short_slug": "esm-c",
  "names": {
   "display": "ESM C",
   "full": "ESM Cambrian",
   "full_null_reason": null,
   "aliases": [
    "ESM C",
    "ESMC",
    "ESM Cambrian",
    "esmc-300m",
    "esmc-600m",
    "ESMC 6B"
   ],
   "ambiguous_aliases": [
    "ESM",
    "ESM2"
   ]
  },
  "family": null,
  "family_null_reason": "300M/600M/6B are size variants of one model and the 2026-05 refresh is recorded as a milestone and note, not a family",
  "domain": "protein",
  "domains_secondary": [],
  "one_liner": "Sequence-only protein language model positioned as a replacement for ESM-2",
  "org": {
   "names": [
    "Evolutionary Scale",
    "Chan Zuckerberg Biohub"
   ],
   "global_entity_ids": [
    "evolutionary-scale"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "pLM (representation)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT plus third-party notice (per current blog and model card)",
   "code_license": "SDK on a public repository",
   "license_changed": null,
   "note": "The current blog and model card state that 300M/600M/6B are all open weights under MIT for academic and commercial use; check the third-party notice first"
  },
  "first_release": {
   "date": "2024-12-04",
   "precision": "day",
   "event_kind": "new-weights",
   "basis": "ESM Cambrian blog, with the 300M/600M weights first committed to a model host the same day",
   "timestamp_utc": null,
   "source_url": "https://www.evolutionaryscale.ai/blog/esm-cambrian",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-12-04",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Blog plus the 300M/600M weights, the earliest provable public evidence",
    "timestamp_utc": null,
    "source_url": "https://www.evolutionaryscale.ai/blog/esm-cambrian",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-06-03",
    "precision": "day",
    "event_kind": "paper-revision",
    "basis": "Protein world-model preprint released with the Biohub-era generation",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.06.03.729735",
    "verification": "partial",
    "note": "The preprint full text was unreachable this round; only the record and official pages were used"
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2026-05-27",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights axis: updated ESMC weights and SAE models; the 6B model opened separately on 2025-12-01 and does not belong to this event",
    "timestamp_utc": null,
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-05-27",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "The institute announced ESMC, ESMFold2 and a sequence atlas as a world model of protein biology",
    "timestamp_utc": null,
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-08-27",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "esm SDK v3.4.0 added full ESMC support (SDK and weights are two separate axes)",
   "timestamp_utc": null,
   "source_url": "https://github.com/evolutionaryscale/esm/releases",
   "verification": "verified",
   "note": "Later SDK releases v3.4.1-v3.4.3 run to 2026-11-09; the latest weights-axis event is 2026-05-27"
  },
  "status": "active",
  "status_note": "Both the SDK and weights axes had events within 12 months",
  "what_it_does": "A sequence-only representation model for embeddings, mutation-effect scoring and downstream supervised tasks, a different job from the multimodal generation line of ESM3. The updated 6B version has 80 layers and is frozen to serve as the backbone of ESMFold2.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "UniRef, MGnify and JGI clustered at 70% identity into about 83M, 372M and 2B clusters; the institutional page separately states about 2.8B training sequences (developer framing)",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm-cambrian"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "The official pages list only those public databases and disclose no private data",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm-cambrian"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Five-target binder experiments in the updated release, with no claim of feeding results back into training; denominators and attribution unverified",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Five-target binder validation (unverified)",
    "modality": "mini-binder; antibody-derived",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Unverified: denominators, the hit definition and whether rates are per design or per target are all undisclosed and the preprint full text was rate-limited, so no figures are given",
    "success_rate": null,
    "affinity_best": null,
    "version": "institutional news plus shared preprint",
    "source_type": "self",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "date": {
     "date": "2026-05-27",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "Announced with the updated release",
     "timestamp_utc": null,
     "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
     "verification": "partial",
     "note": null
    },
    "verification": "unverified",
    "note": "Unverified: the campaign is shared across several models in the same release and is not an ESM C-specific result; the two shared rate ranges also attributed to ESMFold2 are not adopted this round."
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "DMS substitutions zero-shot, maintainer-run summary table (updated 2025-06-13); 300M and 600M entries",
    "source_url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "confidence": "high",
    "verification": "verified",
    "note": "Both the 300M and 600M entries were maintainer-run; no separate entry for the updated weights was found"
   }
  ],
  "other_evaluations": [
   {
    "label": "Size-matched comparison against ESM-2",
    "category": "comparison-vs-baseline-only",
    "detail": "The official blog says the 300M model matches ESM-2 650M and the 600M model compares with ESM-2 3B (developer self-report)",
    "source_type": "self"
   },
   {
    "label": "Binder design wet lab (unverified)",
    "category": "wet-lab-validation",
    "detail": "Five-target binder experiments in the updated release; total design counts undisclosed, so no figures",
    "source_type": "self"
   },
   {
    "label": "Structure boards entered via a downstream model",
    "category": "related-model-evaluation",
    "detail": "On structure boards the entrant is ESMFold2, built on the frozen 6B; ESM C itself does not report there",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: better representation quality per unit cost than ESM-2, with 300M matching ESM-2 650M and 600M matching ESM-2 3B (self-reported, no third-party check found).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Both ProteinGym entries were run by the benchmark maintainers; no independent evidence exists for the wet-lab side.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Evolutionary Scale was founded in 2023-07 by the former Meta FAIR ESM team as a public benefit corporation; in 2025-11 the team joined Chan Zuckerberg Biohub on undisclosed terms and the model repositories moved to that organization.",
   "funding": "Seed round of $142M on 2024-06-25 led by Lux Capital with two angel investors, joined by Amazon and NVentures; valuation undisclosed, as are the Biohub transaction terms.",
   "funding_null_reason": null,
   "products_partnerships": "Siblings are ESM3 and ESMFold2; on 2026-05-27 ESMC, ESMFold2 and a sequence atlas were announced together as a world model. Competing lines include ProGen3 and Profluent-E1.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-esm3",
    "model-esmfold2",
    "model-progen3",
    "model-profluent-e1"
   ]
  },
  "reading_notes": [
   "The 300M and 600M entries sit almost level on the board: doubling size buys nothing on zero-shot mutation effects.",
   "The SDK and the weights are two version axes and should be read separately; the 6B model opened on 2025-12-01, not in the 2026-05-27 event.",
   "Licensing follows the current blog and model card (MIT plus a third-party notice); check that notice before use.",
   "The binder rates are developer framing from a campaign shared across models; no figures are given this round and they cannot be compared with design models."
  ],
  "caveats": [
   "The two binder rate ranges shared with ESMFold2 are unverifiable; the preprint full text was rate-limited.",
   "The latest update is an SDK event, not new weights; the older record on the site conflated the two.",
   "The figure of about 2.8B sequences appears only on the institutional page, uses a different basis from the cluster counts and must not be added to them.",
   "Per this round's version guard, 'ESM2' must not link to this record."
  ],
  "sources": [
   {
    "url": "https://www.evolutionaryscale.ai/blog/esm-cambrian",
    "title": "ESM Cambrian official blog",
    "kind": "company-blog",
    "date": {
     "date": "2024-12-04",
     "precision": "day",
     "event_kind": "new-weights",
     "basis": "first release",
     "timestamp_utc": null,
     "source_url": "https://www.evolutionaryscale.ai/blog/esm-cambrian",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://huggingface.co/EvolutionaryScale/esmc-600m-2024-12",
    "title": "ESM C 600M model card",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://huggingface.co/biohub/ESMC-6B",
    "title": "Updated ESMC 6B model card",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://biohub.org/news/world-model-of-protein-biology/",
    "title": "Institutional world-model release page",
    "kind": "official-site",
    "date": {
     "date": "2026-05-27",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "release day",
     "timestamp_utc": null,
     "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://www.biorxiv.org/content/10.64898/2026.06.03.729735",
    "title": "Protein world-model preprint (full text not retrieved)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/evolutionaryscale/esm/releases",
    "title": "esm SDK releases",
    "kind": "repository",
    "date": {
     "date": "2026-08-27",
     "precision": "day",
     "event_kind": "code-release",
     "basis": "v3.4.0",
     "timestamp_utc": null,
     "source_url": "https://github.com/evolutionaryscale/esm/releases",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "ProteinGym official summary table",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-05-27 (updated weights)",
    "now": "2026-08-27 (SDK v3.4.0 added full ESMC support)",
    "source_url": "https://github.com/evolutionaryscale/esm/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "new weights",
    "now": "code release; the SDK and weights are recorded as two axes",
    "source_url": "https://github.com/evolutionaryscale/esm/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "release_milestones.weights_release",
    "was": "The 6B model opened with the 2026-05-27 update",
    "now": "The 6B model opened on 2025-12-01 and is not part of the 2026-05-27 event",
    "source_url": "https://huggingface.co/biohub/ESMC-6B",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation[0]",
    "was": "The two binder rate ranges on five targets were published as confirmed figures",
    "now": "No figures: the full text was rate-limited and the campaign is shared across models, so it is marked unverified",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "date_found": "2026-11-16"
   },
   {
    "field": "open_status",
    "was": "Early framing: 6B via API only, with non-commercial terms for 300M/600M",
    "now": "The current blog and model card state all three sizes are open weights under MIT plus a third-party notice",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm-cambrian",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[0]",
   "release_milestones.paper_v1",
   "training_data.a_public"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-esm-c/",
   "en": "/bio/benchmark/models/esm-c/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-esm3",
  "kind": "model",
  "short_slug": "esm3",
  "names": {
   "display": "ESM3",
   "full": null,
   "full_null_reason": "Official materials use ESM3 only, with no expanded name",
   "aliases": [
    "ESM3",
    "esm3-sm-open-v1",
    "ESM-3"
   ],
   "ambiguous_aliases": [
    "ESM",
    "ESM2"
   ]
  },
  "family": null,
  "family_null_reason": "The open small model and the larger API version are size variants of one model, not a family",
  "domain": "protein",
  "domains_secondary": [
   "generation",
   "structure"
  ],
  "one_liner": "Multimodal generative protein language model; sequence, structure and function prompt each other",
  "org": {
   "names": [
    "Evolutionary Scale",
    "Chan Zuckerberg Biohub"
   ],
   "global_entity_ids": [
    "evolutionary-scale"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "pLM / design (multimodal generation)",
  "open_status": {
   "code": "open",
   "weights": "research-only",
   "weights_license": "open small model (esm3-sm-open-v1) under non-commercial terms; larger models via API",
   "code_license": "SDK on a public repository",
   "license_changed": null,
   "note": "Small-model weights are downloadable while larger models come only through the API and partners; the two tiers differ and citations must say which was used"
  },
  "first_release": {
   "date": "2024-06-25",
   "precision": "day",
   "event_kind": "new-weights",
   "basis": "Developer ESM3 release blog, with esm3-sm-open-v1 weights posted on a model host the same day",
   "timestamp_utc": null,
   "source_url": "https://www.evolutionaryscale.ai/blog/esm3-release",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-06-25",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Release blog and open weights, the earliest provable public evidence",
    "timestamp_utc": null,
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3-release",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-07-02",
    "precision": "day",
    "event_kind": "paper-revision",
    "basis": "bioRxiv preprint (2024.07.01.600583)",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.07.01.600583v2.full",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-01-16",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Science online (issue 2025-02-21)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1126/science.ads0018",
    "verification": "unverified",
    "note": "Day-level online date unverified: the registry source has no published-online field, only a created date and the print issue"
   },
   "code_release": null,
   "weights_release": {
    "date": "2024-06-25",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Open small-model weights posted with the release blog",
    "timestamp_utc": null,
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3-release",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-06-25",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Developer release blog",
    "timestamp_utc": null,
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2025-01-16",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Science publication (online 2025-01, issue 2025-02-21) plus the API public beta",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1126/science.ads0018",
   "verification": "partial",
   "note": "Day-level date unverified; see the journal_online note"
  },
  "status": "maintained",
  "status_note": "The SDK is updated through v3.4.x, with no new ESM3 weights",
  "what_it_does": "Tokenizes sequence, three-dimensional structure and function annotations and runs masked generation in one transformer, so any modality can prompt any other; the pitch is programmable protein design. That is a different job from its representation-only sibling ESM C.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "UniRef, MGnify, JGI, OAS, PDB, AFDB and others; 2.78B natural proteins for the largest model, or 3.15B sequences, 236M structures and 539M function-annotated proteins once predicted and synthetic data are included",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.07.01.600583v2.full"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "All listed sources are public databases; no licensed private data disclosed",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.07.01.600583v2.full"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "The esmGFP designs were wet-lab tested, with no claim of feeding results back into training",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "esmGFP fluorescent-protein design",
    "modality": "fluorescent protein",
    "targets": null,
    "designs_tested": 96,
    "hits": null,
    "denominator_basis": "Two rounds of 96 generated designs each were tested; the blog gives no hit count, so no hit rate is stated",
    "success_rate": null,
    "affinity_best": null,
    "version": "developer blog",
    "source_type": "self",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3",
    "date": {
     "date": "2024-06-25",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "Published with the release blog",
     "timestamp_utc": null,
     "source_url": "https://www.evolutionaryscale.ai/blog/esm3",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "note": "The site said 'synthesized' where it should say 'tested'; round two produced designs as bright as the natural protein. Developer-run, not an independent replication."
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "DMS substitutions zero-shot, maintainer-run summary table (updated 2025-06-13); open 1.4B entry",
    "source_url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "confidence": "high",
    "verification": "verified",
    "note": "The entry was run by the benchmark maintainers on the open small model and does not stand for the larger API version"
   },
   {
    "benchmark_id": "bench-40-motifbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "MotifBench submission leaderboard (2025-05 submission)",
    "source_url": "https://github.com/blt2114/MotifBench",
    "confidence": "high",
    "verification": "verified",
    "note": null
   },
   {
    "benchmark_id": "bench-34-abbibench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "benchmark-registered entry",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page with no primary document opened, so it displays as unverified"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "self",
    "counted": false,
    "eval_set": "CASP14 (71 targets) and CASP15 (70 targets) time-held-out structure and contact prediction",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.07.01.600583v2.full",
    "confidence": "high",
    "verification": "verified",
    "note": "Post-hoc reuse of a time split, not participation; the paper does not report CASP16"
   }
  ],
  "other_evaluations": [
   {
    "label": "esmGFP design validation",
    "category": "wet-lab-validation",
    "detail": "Wet-lab testing of two rounds of 96 designs; the developers' flagship case, outside the site's 44 benchmarks",
    "source_type": "self"
   },
   {
    "label": "Viral-protein weakness noted by maintainers",
    "category": "third-party-commentary",
    "detail": "Benchmark maintainers note that the ESM line degrades markedly on viral proteins and that zero-shot ability plateaus beyond roughly 1B-4B parameters",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: prompting across all three modalities yields genuinely new functional proteins, demonstrated by esmGFP at about 58% sequence identity to the nearest known fluorescent protein (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The ProteinGym and MotifBench entries were run by the organizers; no independent replication of the wet lab was found.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Evolutionary Scale was founded in 2023-07 by the former Meta FAIR ESM team (which left in 2023-04) as a public benefit corporation, about 20 people at funding; in 2025-11 the team joined Chan Zuckerberg Biohub on undisclosed terms.",
   "funding": "Seed round of $142M on 2024-06-25 led by Lux Capital with two angel investors, joined by Amazon and NVentures (NVIDIA's venture arm); valuation undisclosed.",
   "funding_null_reason": null,
   "products_partnerships": "Siblings are ESM C and ESMFold2, with access also via API; the Biohub that absorbed the team says it will scale to 10,000 GPUs by 2028. Competing lines include ProGen3 and Profluent-E1.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-esm-c",
    "model-esmfold2",
    "model-progen3",
    "model-profluent-e1"
   ]
  },
  "reading_notes": [
   "The board entry is the maintainer-run open 1.4B model, not the larger API model; cite the summary table's date.",
   "esmGFP shows a functional protein can be generated, but one case is not a hit rate, and the blog gives no hit count.",
   "The paper reports only post-hoc CASP14/15 time splits; the claim of better CASP16 results has no primary source and should not be cited.",
   "The AbBiBench row rests on the site's own page and should be read as unverified."
  ],
  "caveats": [
   "The Science online date is day-level but unverified; only a created date and the print issue are checkable.",
   "The two rounds of 96 esmGFP designs give a denominator but no hit count, so any hit rate is extrapolation.",
   "The open small model and the API-only large model differ in licence and do not stand in for each other.",
   "Per this round's version guard, 'ESM2' must not link to this record."
  ],
  "sources": [
   {
    "url": "https://www.evolutionaryscale.ai/blog/esm3",
    "title": "ESM3 release blog",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.evolutionaryscale.ai/blog/esm3-release",
    "title": "ESM3 release note (weights posted)",
    "kind": "company-blog",
    "date": {
     "date": "2024-06-25",
     "precision": "day",
     "event_kind": "new-weights",
     "basis": "release day",
     "timestamp_utc": null,
     "source_url": "https://www.evolutionaryscale.ai/blog/esm3-release",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://doi.org/10.1126/science.ads0018",
    "title": "ESM3 journal paper",
    "kind": "journal",
    "date": {
     "date": "2025-01-16",
     "precision": "day",
     "event_kind": "journal-publication",
     "basis": "online date unverified",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1126/science.ads0018",
     "verification": "unverified",
     "note": null
    }
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2024.07.01.600583v2.full",
    "title": "ESM3 preprint full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "ProteinGym official summary table",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://github.com/blt2114/MotifBench",
    "title": "MotifBench repository and leaderboard",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://github.com/evolutionaryscale/esm/releases",
    "title": "esm SDK releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://techcrunch.com/2024/06/25/evolutionaryscale-backed-by-amazon-and-nvidia-raises-142m-for-protein-generating-ai",
    "title": "Seed-round report",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.gunder.com/en/news-insights/client-news/gunderson-client-evolutionaryscale-team-joins-biohub-in-strategic-transaction-with-chan-zuckerberg-initiative",
    "title": "Announcement of the team joining Biohub",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation[0]",
    "was": "Two rounds of 96 designs each 'synthesized'",
    "now": "Two rounds of 96 designs each 'tested'; the blog gives no hit count, so no hit rate",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "The Science online date was treated as confirmed",
    "now": "The day-level online date is marked unverified; only a created date and the print issue are checkable",
    "source_url": "https://doi.org/10.1126/science.ads0018",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update.date",
   "release_milestones.journal_online.date",
   "benchmarks_reported[2]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-esm3/",
   "en": "/bio/benchmark/models/esm3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-esmfold2",
  "kind": "model",
  "short_slug": "esmfold2",
  "names": {
   "display": "ESMFold2",
   "full": null,
   "full_null_reason": "Official materials use ESMFold2 only; no expanded name",
   "aliases": [
    "ESMFold2",
    "ESMFold2-Fast",
    "ESMFold2 Pro"
   ],
   "ambiguous_aliases": [
    "ESMFold",
    "ESM"
   ]
  },
  "family": null,
  "family_null_reason": "Covers ESMFold2 with its Fast and Pro variants; ESM C and the ESM Atlas are separate entries",
  "domain": "structure",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Protein-language-model co-folding that needs no MSA retrieval, released alongside a sequence atlas",
  "org": {
   "names": [
    "Biohub",
    "EvolutionaryScale team"
   ],
   "global_entity_ids": [
    "evolutionary-scale"
   ]
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding (single-sequence PLM)",
  "open_status": {
   "code": "unclear",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Weights are public on Hugging Face, but licence metadata is inconsistent across the organisation's model cards, so neither term is settled"
  },
  "first_release": {
   "date": "2026-05-27",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Research-institute release page (HF repos created 2026-05-13, not used as first release)",
   "timestamp_utc": null,
   "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-05-27",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Institute release page",
    "timestamp_utc": null,
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2026-05-27",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights public on Hugging Face at release",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/biohub/ESMFold2",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-05-27",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Institute release page, published with ESM C and the ESM Atlas",
    "timestamp_utc": null,
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-11-09",
   "precision": "day",
   "event_kind": "patch",
   "basis": "esm SDK v3.4.3",
   "description": "v3.4.3 is bug fixes plus a transformers dependency bump; the new weights belong to the previous release v3.4.2 (2026-10-30, with the 2025-cutoff weights and the Pro folding API), and HF weights were updated 2026-09-14",
   "timestamp_utc": null,
   "source_url": "https://github.com/evolutionaryscale/esm/releases",
   "verification": "verified",
   "note": "Read the SDK and weights axes separately"
  },
  "status": "active",
  "status_note": "The SDK still ships patch releases",
  "what_it_does": "Puts evolutionary information inside a pretrained protein language model, predicts multi-chain complexes without MSA retrieval at inference, and ships a binder design pipeline. Unlike AlphaFold3 and Boltz-2 it needs no MSA and runs faster; ESM3 and ESM C are its siblings.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB plus AFDB (cut-off 2021-09), with the ESM C sequence corpus for the language model",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No claim that wet-lab results were fed back into training",
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Binder designs for five targets were assayed in the wet lab (unverified)",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Binder campaign shared across the release",
    "modality": "mini-binder; antibody-derived",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Denominators undisclosed; the preprint full text is rate-limited and the campaign is shared across several models in one release, so it cannot be attributed to a single model (unverified)",
    "success_rate": null,
    "affinity_best": null,
    "version": "institutional news plus shared preprint",
    "source_type": "self",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "date": {
     "date": "2026-05-27",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "Institute release page",
     "timestamp_utc": null,
     "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
     "verification": "verified",
     "note": null
    },
    "verification": "unverified",
    "note": "Not an ESMFold2-specific wet-lab paper and not an independent replication; no hit-rate figures are given (unverified)"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "FoldBench no-MSA complexes / antibody-antigen",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "confidence": "medium",
    "verification": "partial",
    "note": "Model-card figures reach the site second-hand; a competitor paper reports another set on a different antibody subset"
   },
   {
    "benchmark_id": "bench-34-abbibench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "AbBiBench structure-prediction entry",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified), and such entries are paper-figure correlations registered by the benchmark, not submissions or reruns"
   }
  ],
  "other_evaluations": [
   {
    "label": "Sequence and structure atlas released alongside",
    "category": "resource-release",
    "detail": "The release includes an atlas of 6.8B sequences and 1.1B predicted structures; these are official scale figures, not benchmark results",
    "source_type": "self"
   },
   {
    "label": "Competitor-run evaluation",
    "category": "competitor-reported",
    "detail": "The OpenDDE report evaluates this model on a self-defined antibody subset under a different protocol, with no independent replication",
    "source_type": "third"
   },
   {
    "label": "Hosting and second-hand accounts",
    "category": "second-hand",
    "detail": "A third-party platform hosts the model, and its card figures mostly reach readers through platform and community write-ups (second-hand)",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report matching or beating AlphaFold3 and Protenix-v1 on protein-protein and antibody-antigen targets without MSAs (self-reported, no independent replication).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Both figure sets are self-reported, by developer and competitor; neither was replicated",
  "background": {
   "org_profile": "Biohub is a philanthropically funded nonprofit medical research organization that consolidated related biomedical research in 2025-11; the EvolutionaryScale team joined on 2025-11-06 in a strategic transaction with undisclosed terms.",
   "funding": null,
   "funding_null_reason": "Nonprofit funded by its philanthropic backer; amounts undisclosed",
   "products_partnerships": "Released 2026-05-27 together with ESM C and the atlas and hosted free on the organisation's own platform; the developers also ran the pipeline to design binders for five cancer and immunology targets.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-esm3",
    "model-esm-c",
    "model-alphafold3",
    "model-boltz-2"
   ]
  },
  "reading_notes": [
   "Self-reported numbers sit on FoldBench and are not comparable with PoseBusters-style protocols.",
   "The antibody-antigen claim exists in two different protocols, one from the developer and one from a competitor.",
   "The real value of dropping MSAs is speed and ease of use; the accuracy claim still awaits independent evaluation.",
   "The design-validation denominators are undisclosed, so hit rates cannot be cited."
  ],
  "caveats": [
   "The wet-lab campaign is shared across the release's models and the preprint is rate-limited, so no figures are given (unverified).",
   "The Runs N' Poses claim has no matching figure in official material and is treated as unconfirmed.",
   "Licence metadata is inconsistent across the organisation's models, so openness cannot be settled.",
   "The AbBiBench line rests on the site's own page and is shown as unverified."
  ],
  "sources": [
   {
    "url": "https://biohub.org/news/world-model-of-protein-biology/",
    "title": "Institute release page: a world model of protein biology",
    "kind": "official-site",
    "date": {
     "date": "2026-05-27",
     "precision": "day",
     "event_kind": "product-update",
     "basis": "Release date",
     "timestamp_utc": null,
     "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://huggingface.co/biohub/ESMFold2",
    "title": "Hugging Face model card",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/evolutionaryscale/esm/releases",
    "title": "esm SDK releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://rosettacommons.org/2026/06/09/biohub-releases-open-source-models-that-utilize-evolution-to-design-disease-target-binders/",
    "title": "Community write-up (second-hand)",
    "kind": "second-hand",
    "date": null
   },
   {
    "url": "https://www.gunder.com/en/news-insights/client-news/gunderson-client-evolutionaryscale-team-joins-biohub-in-strategic-transaction-with-chan-zuckerberg-initiative",
    "title": "Announcement of the team joining",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-10-30",
    "now": "2026-11-09 (v3.4.3 is the later release)",
    "source_url": "https://github.com/evolutionaryscale/esm/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "new weights",
    "now": "Patch: v3.4.3 is only bug and dependency fixes; the new weights belong to the 2026-10-30 release",
    "source_url": "https://github.com/evolutionaryscale/esm/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "hit-rate ranges for minibinders and antibody-derived formats",
    "now": "No figures: the preprint is rate-limited and the campaign is shared across the release's models (unverified)",
    "source_url": "https://biohub.org/news/world-model-of-protein-biology/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[0].success_rate",
   "benchmarks_reported[1].source_url",
   "open_status.code"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-esmfold2/",
   "en": "/bio/benchmark/models/esmfold2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-ether0",
  "kind": "model",
  "short_slug": "ether0",
  "names": {
   "display": "ether0",
   "full": null,
   "aliases": [
    "ether0"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single weights release with no family of versions",
  "domain": "small-molecule",
  "domains_secondary": [
   "text-agents"
  ],
  "one_liner": "A 24B-parameter chemistry reasoning model whose answers are molecules",
  "org": {
   "names": [
    "FutureHouse"
   ],
   "global_entity_ids": [
    "futurehouse"
   ]
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Chemistry reasoning LLM (reinforcement-learning fine-tune)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": null,
   "license_changed": null,
   "note": "Weights under Apache-2.0, with the self-built evaluation set public as well"
  },
  "first_release": {
   "date": "2025-06-04",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1, with weights and the release post alongside",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2506.17238",
   "verification": "verified",
   "note": "The developer blog is one day later and does not move the first-release date"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-06-04",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2506.17238",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-06-04",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2506.17238",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-10-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Last maintenance commit in the repository",
    "event_kind": "code-maintenance-only",
    "timestamp_utc": null,
    "source_url": "https://github.com/Future-House/ether0/commits/main.atom",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-06-04",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights repository made public",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/futurehouse/ether0",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-06-05",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer release post",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.futurehouse.org/news/ether0-a-scientific-reasoning-model-for-chemistry",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2025-12-02",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v2 revision",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2506.17238",
   "verification": "verified",
   "description": "The preprint reached v2 on 2 December 2025; earlier events are a repository maintenance commit on 26 October 2025 and acceptance at a conference that year.",
   "note": "The site previously recorded the 2025-10-26 maintenance commit as the latest update"
  },
  "status": "active",
  "status_note": "A paper revision and maintenance commits within 12 months",
  "what_it_does": "Recasts chemistry tasks as verifiable questions whose answer is a molecule, then applies supervised fine-tuning and reinforcement learning to a 24-billion-parameter open general model, spanning solubility, functional groups, reaction prediction, absorption and metabolism, acid dissociation and hundreds of other task types. Unlike instruction-tuned models on public therapeutics datasets such as TxGemma, it follows a reasoning-with-verifiable-rewards route.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Several public chemistry and pharmacology measurement archives, with questions derived from existing measurements; the paper reports 375 task types and 640,730 questions",
    "source_url": "https://arxiv.org/abs/2506.17238"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No new experiments; questions come from existing measurements",
    "source_url": "https://arxiv.org/abs/2506.17238"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-07-uspto",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2506.17238",
    "confidence": "high",
    "second_hand": false,
    "note": "A self-built reaction-prediction task and held-out set rather than that dataset's official test split, with the comparison model trained at a different scale"
   }
  ],
  "other_evaluations": [
   {
    "label": "Self-built chemistry question set",
    "category": "own-task-set",
    "detail": "A public evaluation set of 25 questions per task type, released alongside the model",
    "source_type": "self"
   },
   {
    "label": "Multiple-choice comparison with a frontier general model",
    "category": "comparison-vs-baseline-only",
    "detail": "Reported side by side with a closed reasoning model on zero-shot multiple choice",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a specialist chemistry model trained with verifiable rewards that beats a classical reaction model on a self-built held-out task using far fewer samples, with weights and question set fully open.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation was found, though open weights and an open question set make one possible",
  "background": {
   "org_profile": "The developer is a non-profit research institute founded in 2023 in San Francisco, funded mainly by one technology-sector donor, producing research agents and open models; a for-profit spinout followed in November 2025.",
   "funding": "The non-profit runs on donations; the spinout announced a $70M seed round in December 2025 at a reported valuation of about $250M, co-led by two venture firms.",
   "products_partnerships": "The same institute produces literature and experiment agents; related records here include Robin and Kosmos.",
   "related_models": [
    "model-txgemma",
    "model-robin",
    "model-kosmos"
   ]
  },
  "reading_notes": [
   "The reaction comparison differs in both training size and task construction and cannot be read as a standing on that dataset's official board.",
   "The question set is defined by the developer, so the conclusions cover only chemistry tasks inside that training distribution.",
   "Zero-shot multiple choice sits slightly below a closed reasoning model, showing the gap that remains between a specialist small model and frontier general ones.",
   "Open weights and an open question set are its main transparency advantage, since third parties can recheck directly."
  ],
  "caveats": [
   "Its only pairing here uses a self-built task, not the official test split.",
   "It reports no score on the common small-molecule property boards.",
   "Questions derive from existing measurements; there is no new experimental data.",
   "The latest update is a paper revision; the weights have not been re-released."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2506.17238",
    "title": "Preprint 2025",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/futurehouse/ether0",
    "title": "Weights repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.futurehouse.org/news/ether0-a-scientific-reasoning-model-for-chemistry",
    "title": "Developer release post",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://github.com/Future-House/ether0/commits/main.atom",
    "title": "Repository commit feed",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2025-10-26 (repository maintenance commit)",
    "now": "2025-12-02 (preprint v2 revision)",
    "source_url": "https://arxiv.org/abs/2506.17238",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.weights_release",
   "release_milestones.official_announcement",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": "futurehouse",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-ether0/",
   "en": "/bio/benchmark/models/ether0/"
  },
  "names_full_null_reason": "The model carries a lowercase code name with no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-evo-2",
  "kind": "model",
  "short_slug": "evo-2",
  "names": {
   "display": "Evo 2",
   "full": null,
   "aliases": [
    "Evo 2",
    "Evo2",
    "Evo2-7B",
    "Evo2-40B"
   ],
   "ambiguous_aliases": [
    "Evo"
   ]
  },
  "family": null,
  "family_null_reason": "7B and 40B are sizes of one model, not family members; the predecessor Evo has no record",
  "domain": "genomics-rna",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Open DNA language model with million-token context for zero-shot variant scoring and sequence generation",
  "org": {
   "names": [
    "Arc Institute",
    "NVIDIA",
    "Stanford University",
    "University of California, Berkeley",
    "University of California, San Francisco"
   ],
   "global_entity_ids": [
    "arc-institute"
   ]
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Genome language model (sequence generation + variant effects)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Weights, training and inference code and the pretraining dataset are all public and integrated into a chip vendor's bio framework; licence text not retrieved this round"
  },
  "first_release": {
   "date": "2025-02-19",
   "precision": "day",
   "year_hint": null,
   "basis": "Joint launch by the institute and a chip vendor (preprint the same month)",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://blogs.nvidia.com/blog/evo-2-biomolecular-ai/",
   "verification": "partial",
   "note": "The launch date comes from coverage rather than a primary page; the code and weights repositories were created earlier and repository creation is not used as first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-02-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Launch coverage with weights available",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://blogs.nvidia.com/blog/evo-2-biomolecular-ai/",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-02",
    "precision": "month",
    "year_hint": null,
    "basis": "Preprint (month-level evidence only)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13128491/",
    "verification": "partial",
    "note": "Not padded to a day"
   },
   "journal_online": {
    "date": "2026-03-04",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal online (received 2025-02-18, accepted 2026-01-22)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41586-026-10176-5",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2026-02-28",
    "precision": "day",
    "year_hint": null,
    "basis": "GitHub v0.5.0",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/ArcInstitute/evo2/releases",
    "verification": "verified",
    "note": "Public code predates this; v0.5.0 is the version anchor retrieved this round"
   },
   "weights_release": {
    "date": "2025-02-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights public at launch",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/arcinstitute/evo2_40b",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-02-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Joint launch blog",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://blogs.nvidia.com/blog/evo-2-biomolecular-ai/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-03-04",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1038/s41586-026-10176-5",
   "verification": "verified",
   "description": "Journal online publication; GitHub v0.5.0 came just before (2026-02-28) and later commits (kernel work, 2026-06-19) are minor. No official source for a next-generation model was found as of the as_of date",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication and a version release within 12 months",
  "what_it_does": "Autoregressively pretrained on DNA with context up to a million nucleotides across bacteria, archaea, eukaryotes and phages; scores variants zero-shot without functional labels and generates new sequences for design. Viruses infecting humans and complex eukaryotes were excluded from training. Unlike AlphaGenome it is self-supervised rather than supervised on functional tracks; like NTv3 it is a long-context genome model, with broader species coverage.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "A public pan-genome dataset (prokaryotic, eukaryotic and phage sequences); token counts and nucleotide counts are kept as two separate framings per the paper",
    "source_url": "https://huggingface.co/datasets/arcinstitute/opengenome2"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13128491/"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Designed sequences integrated into cells and checked by chromatin-accessibility sequencing, for validation only",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13128491/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Generated-DNA accessibility validation",
    "targets": null,
    "designs_tested": 36,
    "hits": 33,
    "denominator_basis": "36 constructs (5 in one cell line, 31 in another), per the figure legend",
    "success_rate": "33/36; of 24 designs requiring differential accessibility, 4/24 exceeded twofold and 1/24 threefold",
    "affinity_best": null,
    "version": "journal paper",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13128491/",
    "date": {
     "date": "2026-03-04",
     "precision": "day",
     "year_hint": null,
     "basis": "Journal online publication",
     "event_kind": "journal-publication",
     "timestamp_utc": null,
     "source_url": "https://api.crossref.org/works/10.1038/s41586-026-10176-5",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "New experiments by the developing institute; the method paper is also the wet-lab paper",
    "unverified_marker": null,
    "note": "Success is defined by accessibility matching the designed pattern, not a functional phenotype"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-11-clinvar",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13128491/",
    "confidence": "medium",
    "second_hand": false,
    "note": "Developer-defined coding and non-coding subsets with values only in figures; the text gives rank-style statements"
   },
   {
    "benchmark_id": "bench-37-traitgym",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://www.biorxiv.org/content/10.1101/2025.02.11.637758",
    "confidence": "high",
    "second_hand": false,
    "note": "Run by the benchmark's own authors under their protocol; not reported in this model's paper"
   },
   {
    "benchmark_id": "bench-39-rnagym",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://arxiv.org/pdf/2608.22849",
    "confidence": "medium",
    "second_hand": true,
    "note": "The matrix marks this as a board entry, but only a figure quoted in another model's paper exists; no row on the model page and the board entry is unverified (kept as an open inconsistency)"
   }
  ],
  "other_evaluations": [
   {
    "label": "Single-gene saturation mutagenesis classification",
    "category": "own-task-set",
    "detail": "A supervised classifier on the model's representations beats its own zero-shot scores; a developer-defined setup",
    "source_type": "self"
   },
   {
    "label": "Third-party controlled mitochondrial and nuclear variant study",
    "category": "comparison-vs-baseline-only",
    "detail": "A third-party preprint says the model calls pathogenic variants benign when only flanking sequence changes, arguing sequence regularity drives predictions; the abstract gives no metric. A comparison figure once cited on the site had no locatable source and is removed",
    "source_type": "third"
   },
   {
    "label": "Third-party probes on frozen embeddings",
    "category": "comparison-vs-baseline-only",
    "detail": "A third party trained covariance probes on frozen embeddings and claims parity or better with dedicated predictors; these are supervised probes, not zero-shot, and must not be read with the table above. The competitor list once paraphrased is not in the abstract and is removed",
    "source_type": "third"
   },
   {
    "label": "BEND",
    "category": "public-benchmark-not-on-site",
    "detail": "The original paper does not report this benchmark",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one open all-domains sequence model both scores variants zero-shot and generates designed sequences, with a particular edge on indel-type variants competitors cannot score.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A third party ran the model under the TraitGym protocol, and two further third-party studies note misclassification when only flanking sequence changes and that high probe scores do not equal zero-shot ability.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Led by a California non-profit research institute (founded 2021 with over $650M committed at launch) whose core investigators hold multi-year renewable terms alongside university appointments; no citable official headcount exists, so none is given. Partners are a listed chip company and three US universities.",
   "funding": "Non-profit, endowment-funded; 2024 budget figures differ between sources so no single number is given. Compute came from the partner chip company as 2,000 accelerators on a cloud platform (not \"over 2,000\").",
   "products_partnerships": "Builds on the institute's earlier prokaryote and phage genome model; the institute's single-cell line includes STATE and Stack; comparators are AlphaGenome and NTv3; a third-party company runs probing studies on its frozen embeddings.",
   "related_models": [
    "model-alphagenome",
    "model-ntv3",
    "model-state",
    "model-stack"
   ]
  },
  "reading_notes": [
   "On ClinVar the edge is on indel-type variants competitors cannot score; it is not first on coding SNVs, so \"beats everything\" paraphrases do not hold, and values appear only in figures.",
   "High probe scores on embeddings are not zero-shot ability and must be read separately.",
   "Third parties find dedicated methods stronger on mitochondrial and missense variants, showing a trade-off between all-domains coverage and single-task accuracy.",
   "The RNAGym \"board\" entry is only a figure quoted by another paper and awaits verification."
  ],
  "caveats": [
   "Site benchmark values sit in figures or come from third parties; the text gives no citable numbers.",
   "The RNAGym pairing is unconfirmed and the matrix label disagrees with the benchmark page.",
   "Wet-lab success means the accessibility pattern matched, not a functional phenotype.",
   "The launch date and preprint date rest on coverage and month-level evidence; partly verified."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41586-026-10176-5",
    "title": "Journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13128491/",
    "title": "Journal full text (public archive)",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/ArcInstitute/evo2/releases",
    "title": "Code repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/datasets/arcinstitute/opengenome2",
    "title": "Pretraining dataset card",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.02.11.637758",
    "title": "Third-party benchmark preprint",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://doi.org/10.64898/2026.03.10.710786",
    "title": "Third-party controlled-study preprint",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://doi.org/10.64898/2026.04.10.717844",
    "title": "Third-party probing preprint",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "other_evaluations",
    "was": "Two comparison numbers from a third-party mitochondrial evaluation",
    "now": "Removed: no source can be found on public pages",
    "source_url": "https://doi.org/10.64898/2026.03.10.710786",
    "date_found": "2026-11-16"
   },
   {
    "field": "other_evaluations",
    "was": "Probe study \"beats four named models\"",
    "now": "Removed: the abstract does not list the competitors",
    "source_url": "https://doi.org/10.64898/2026.04.10.717844",
    "date_found": "2026-11-16"
   },
   {
    "field": "training_data.a_public",
    "was": "More than 128,000 genome and metagenome datasets",
    "now": "Removed: neither the paper text nor the dataset card uses this figure",
    "source_url": "https://huggingface.co/datasets/arcinstitute/opengenome2",
    "date_found": "2026-11-16"
   },
   {
    "field": "background.funding",
    "was": "More than 2,000 accelerator cards",
    "now": "2,000 accelerator cards",
    "source_url": "https://blogs.nvidia.com/blog/evo-2-biomolecular-ai/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "release_milestones.weights_release",
   "benchmarks_reported[2]",
   "open_status.code_license",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-evo-2/",
   "en": "/bio/benchmark/models/evo-2/"
  },
  "names_full_null_reason": "The paper uses the model name and sizes only",
  "n_benchmarks_reported": {
   "counted": 2,
   "by_relation": {
    "self-reported-modified-protocol": 1,
    "retrospective-dataset-use": 1,
    "disputed-or-unconfirmed": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developing institute but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-evolvepro",
  "kind": "model",
  "short_slug": "evolvepro",
  "names": {
   "display": "EVOLVEpro",
   "full": null,
   "full_null_reason": "Sources use EVOLVEpro only; no expanded name",
   "aliases": [
    "EVOLVEpro",
    "EvolvePro"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single method; no family boundary needed",
  "domain": "protein",
  "domains_secondary": [
   "antibody"
  ],
  "one_liner": "Few-shot active-learning wrapper for protein engineering, about a dozen variants per round",
  "org": {
   "names": [
    "Brigham and Women's Hospital / Harvard Medical School",
    "Mass General Brigham Gene and Cell Therapy Institute",
    "Beth Israel Deaconess Medical Center",
    "MIT"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (active learning on pLM embeddings)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Two public repositories; it has no weights of its own and relies on third-party pLM weights"
  },
  "first_release": {
   "date": "2024-07-18",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv preprint (date carried from the existing note, not re-fetched this round)",
   "timestamp_utc": null,
   "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
   "verification": "partial",
   "note": "The preprint's first posting date was not re-verified this round"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-07-18",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv preprint (date carried from the existing note, not re-fetched this round)",
    "timestamp_utc": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
    "verification": "partial",
    "note": "The preprint's first posting date was not re-verified this round"
   },
   "paper_v1": {
    "date": "2024-07-18",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv preprint (date carried from the existing note, not re-fetched this round)",
    "timestamp_utc": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
    "verification": "partial",
    "note": "The preprint's first posting date was not re-verified this round"
   },
   "journal_online": {
    "date": "2025-01-24",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Science journal publication (Crossref issued date)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1126/science.adr6006",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-01-24",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Science journal publication (DOI 10.1126/science.adr6006)",
   "description": "Journal version online; repository commits through 2025-06-11 are maintenance.",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1126/science.adr6006",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Latest substantive event is the 2025-01 journal paper; later commits are maintenance",
  "what_it_does": "Uses frozen large protein-language-model embeddings as features and fits a light regressor to each round of measured activity, then picks the next batch of mutants; it pretrains nothing itself. Unlike zero-shot models such as ESM3, it depends on experimental data for each target protein, so it fits settings that already have an activity assay.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Frozen ESM-2 15B weights pretrained on public sequence data; 12 DMS sets only for in silico grid search",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The paper names no private or licensed data",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/"
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "Per-protein active-learning rounds of about 11-16 measured variants train the top regressor",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The same experiments are training input and application result, not validation only",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "REGN10987 antibody engineering (self-reported)",
    "modality": "antibody",
    "targets": 1,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Five active-learning rounds of about 11-16 variants each; the paper gives percentages, not numerators and denominators",
    "success_rate": "hit rate rose from 9.1% to 54.5% over five rounds; ELISA binding improved 63%",
    "affinity_best": "IC50 11.9 nM",
    "version": "PMC preprint version",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
    "date": {
     "date": "2024-07-18",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "bioRxiv preprint (date carried from the existing note, not re-fetched this round)",
     "timestamp_utc": null,
     "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
     "verification": "partial",
     "note": "The preprint's first posting date was not re-verified this round"
    },
    "verification": "verified",
    "note": "Self-reported with no independent replication; endpoints must be named precisely (a reduced-immunogenicity by-product is not protein activity) and the figures come from an early version"
   },
   {
    "campaign": "Enzyme and protein engineering set (unverified)",
    "modality": "protein",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "4-8 rounds of about 12-16 single mutants each; the abstract and the discussion disagree on the fold-improvement range, so no figures are given (unverified)",
    "success_rate": null,
    "affinity_best": null,
    "version": "PMC preprint version",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
    "date": {
     "date": "2024-07-18",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "bioRxiv preprint (date carried from the existing note, not re-fetched this round)",
     "timestamp_utc": null,
     "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
     "verification": "partial",
     "note": "The preprint's first posting date was not re-verified this round"
    },
    "verification": "partial",
    "note": "Unverified: figures must be shown per endpoint and never as one fold-change range; no independent replication"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "In silico DMS active-learning comparison",
    "category": "own-task-set",
    "detail": "Compares eight representations on 12 DMS datasets; frozen pLM embeddings rank best and beat zero-shot prediction (self-reported)",
    "source_type": "self"
   },
   {
    "label": "REGN10987 antibody engineering",
    "category": "wet-lab-validation",
    "detail": "Five active-learning rounds; the hit rate rose and the final product reached nanomolar affinity (self-reported)",
    "source_type": "self"
   },
   {
    "label": "PsaCas12f, PE2, Bxb1 and T7 RNAP engineering",
    "category": "wet-lab-validation",
    "detail": "4-8 rounds of about a dozen single mutants each; the reported fold gains are inconsistent and unverified",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that few-shot active learning on pLM embeddings improved five different proteins within 4-8 rounds of about a dozen mutants each (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party replication found; the authors disclose patents and company roles",
  "background": {
   "org_profile": "An academic team across Harvard-affiliated hospitals (including a gene and cell therapy institute and a virology and vaccine center) and MIT, with collaborators at the University of Tokyo and the Inamori Research Institute; the work spans genome editing, CRISPR diagnostics and protein engineering.",
   "funding": "The acknowledgements list several NIH grants plus foundation and venture funding; the Japanese co-authors are supported by AMED, JSPS and JST.",
   "funding_null_reason": null,
   "products_partnerships": "No product and no pharma partnership; the authors have filed patents on the method without giving numbers. It builds on a third-party open pLM with no affiliation to that model's publisher.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-esm3"
   ]
  },
  "reading_notes": [
   "These are per-protein experimental results and cannot be lined up against zero-shot model scores.",
   "Each fold improvement starts from a different baseline, so they cannot be pooled into one range.",
   "There is no score relation to ProteinGym: its 12 DMS sets are only a data source.",
   "The tiny per-round budget is the point: with an assay in hand, active learning is more practical than zero-shot."
  ],
  "caveats": [
   "Zero pairings: no site benchmark entry exists.",
   "The enzyme and protein engineering set's fold gains are inconsistent and marked unverified.",
   "All results are developer-reported; no independent replication was found.",
   "Its own wet-lab data feeds the training loop, so it is not comparable to validation-only models."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1126/science.adr6006",
    "title": "Science journal paper",
    "kind": "journal",
    "date": {
     "date": "2025-01-24",
     "precision": "day",
     "event_kind": "journal-publication",
     "basis": "Crossref issued",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1126/science.adr6006",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
    "title": "Preprint full text (PMC)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/mat10d/EvolvePro",
    "title": "Official repository (full data)",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/idmjky/EvolvePro",
    "title": "Official repository (deployable model)",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "A single fold-change range summarised the gains across all engineering cases",
    "now": "The abstract and discussion ranges disagree, so figures are shown per endpoint and no single range is given (unverified)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11275896/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[1]",
   "first_release",
   "open_status.weights"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-evolvepro/",
   "en": "/bio/benchmark/models/evolvepro/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-frontier-llms",
  "kind": "model",
  "short_slug": "frontier-llms",
  "names": {
   "display": "Frontier general-purpose LLMs",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [],
   "ambiguous_aliases": [
    "Claude",
    "GPT",
    "Gemini"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Claude 3.5 Sonnet",
     "id_or_null": null,
     "first_release": {
      "date": "2024-06",
      "precision": "month",
      "event_kind": "new-version",
      "basis": "One of the earliest members cited in the LAB-Bench era (not individually re-checked)",
      "timestamp_utc": null,
      "source_url": null,
      "verification": "unverified",
      "note": null
     }
    },
    {
     "name": "GPT-4o",
     "id_or_null": null,
     "first_release": {
      "date": "2024-05",
      "precision": "month",
      "event_kind": "new-version",
      "basis": "One of the earliest members cited in the LAB-Bench era (not individually re-checked)",
      "timestamp_utc": null,
      "source_url": null,
      "verification": "unverified",
      "note": null
     }
    },
    {
     "name": "Claude Fable 5.2 / Opus 5.5 / Sonnet 5.5 / Haiku 5.5",
     "id_or_null": null,
     "first_release": {
      "date": null,
      "precision": null,
      "event_kind": "successor-model",
      "basis": "Listed as current on the developer model page; no release date stated",
      "timestamp_utc": null,
      "source_url": "https://platform.claude.com/docs/en/about-claude/models/overview",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "Category entry for general closed LLMs from Anthropic, OpenAI and Google as test subjects; not one model, no single first release"
  },
  "family_null_reason": null,
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "Category entry for general frontier LLMs (Claude, GPT, Gemini) tested on bio benchmarks",
  "org": {
   "names": [
    "Anthropic",
    "OpenAI",
    "Google DeepMind"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US",
   "GB"
  ],
  "model_class": "General-purpose LLM",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Proprietary models accessed through APIs"
  },
  "first_release": {
   "date": null,
   "precision": null,
   "event_kind": "new-version",
   "basis": "No single first release for a family entry; earliest cited members c. 2024-05/06 (unverified)",
   "timestamp_utc": null,
   "source_url": null,
   "verification": "unverified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": null,
   "precision": null,
   "event_kind": "successor-model",
   "basis": "Current models listed on the developer model page",
   "description": "Developer page lists Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5 as current, with no release dates; OpenAI and Google dates unverified",
   "timestamp_utc": null,
   "source_url": "https://platform.claude.com/docs/en/about-claude/models/overview",
   "verification": "unverified",
   "note": "The earlier 2026-11-09 was back-derived from a retirement floor, not a release date; the four floors differ, so not one release day"
  },
  "status": "active",
  "status_note": "Vendors keep releasing new models",
  "what_it_does": "This site treats general frontier LLMs such as Claude, GPT and Gemini as one category entry, recording how they fare on bio benchmarks such as LAB-Bench and BixBench. They are test subjects, not biology-specific models.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-14-pubmedqa",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": null,
    "source_url": "https://arxiv.org/html/2604.05081",
    "confidence": "medium",
    "verification": "partial",
    "note": "Google self-reports Gemini 3 numbers in the MedGemma 1.5 report; no official board; no vendor Claude/GPT numbers"
   },
   {
    "benchmark_id": "bench-18-labbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "LABBench2",
    "source_url": "https://futurehouse.org/articles/lab-bench-measuring-capabilities-of-language-models-for-biology-research",
    "confidence": "high",
    "verification": "verified",
    "note": "Entries on the LABBench2 official board (family cell)"
   },
   {
    "benchmark_id": "bench-23-bioasq",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": null,
    "source_url": "https://ceur-ws.org/Vol-4283/paper1.pdf",
    "confidence": "high",
    "verification": "unverified",
    "note": "The BioASQ overview ranks systems, not base models"
   },
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "BixBench3",
    "source_url": "https://arxiv.org/html/2503.00096v3",
    "confidence": "high",
    "verification": "verified",
    "note": "Entries on the BixBench3 official board (family cell); question set differs from the original"
   },
   {
    "benchmark_id": "bench-42-biomysterybench",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": null,
    "source_url": "https://www.anthropic.com/research/Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench",
    "confidence": "high",
    "verification": "verified",
    "note": "Anthropic self-reports on its own BioMysteryBench; Vals AI runs a third-party board"
   },
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": null,
    "source_url": null,
    "confidence": "low",
    "verification": "unverified",
    "note": "Second-hand aggregator scores only; family entry, not a model; basis is only this site's own page, unverified"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "Anthropic (founded 2021, San Francisco), OpenAI (founded 2015-12), Google DeepMind (merged 2023-04, Alphabet)",
   "funding": "Anthropic announced a Series H on 2026-05-28 at a $965B post-money valuation; OpenAI funding figures conflict across reports",
   "funding_null_reason": null,
   "products_partnerships": null,
   "products_partnerships_null_reason": "No product or partnership found",
   "related_models": [
    "model-medgemma-med-gemini",
    "model-gpt-rosalind",
    "model-co-scientist",
    "model-claude-protein-design"
   ]
  },
  "reading_notes": [
   "LAB-Bench and BixBench come from Edison, itself a competitor",
   "Scores shift with thinking and tool settings; not comparable across sources"
  ],
  "caveats": [
   "Not a single model: first release and latest update only make sense per member",
   "OpenAI and Google model names and dates not verified from primary pages"
  ],
  "sources": [
   {
    "url": "https://platform.claude.com/docs/en/about-claude/models/overview",
    "title": "Official page platform.claude.com",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://advances.edisonscientific.com/benchmarks/",
    "title": "Official page advances.edisonscientific.com",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2604.09554",
    "title": "arXiv preprint 2604.09554",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.anthropic.com/research/Evaluating-Claude-For-Bioinformatics-With-BioMysteryBench",
    "title": "Anthropic BioMysteryBench article",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.description",
    "was": "Claude Fable 5.2 released \"alongside\" Opus 5.5 etc., dated 2026-11-09",
    "now": "No release date on the developer page; 2026-11-09 was back-derived from a retirement floor and the four models were not released on one day",
    "source_url": "https://platform.claude.com/docs/en/about-claude/models/overview",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "latest_update",
   "family.members",
   "benchmarks_reported[2]",
   "benchmarks_reported[5]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-frontier-llms/",
   "en": "/bio/benchmark/models/frontier-llms/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-genmol",
  "kind": "model",
  "short_slug": "genmol",
  "names": {
   "display": "GenMol",
   "full": null,
   "aliases": [
    "GenMol",
    "NV-GenMol",
    "NV-GenMol-89M-v2",
    "GenMol NIM"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "v1 and v2 are version axes within one record",
  "domain": "generation",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "NVIDIA's masked discrete-diffusion molecule generator covering several generation tasks in one model",
  "org": {
   "names": [
    "NVIDIA"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Molecule generation (discrete diffusion)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "NVIDIA Open Model License",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code Apache-2.0 and weights under the NVIDIA Open Model License; the NIM microservice uses the NVIDIA software licence"
  },
  "first_release": {
   "date": "2025-01-10",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv preprint v1 submitted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2501.06158",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-01-10",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2501.06158",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-01-10",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv preprint v1 (v3 on 2025-07-22)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2501.06158",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2025-01",
    "precision": "month",
    "year_hint": null,
    "basis": "v1 weights on NGC (month-level evidence only)",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://catalog.ngc.nvidia.com/orgs/nvidia/clara/resources/genmol_v1",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-04-24",
   "precision": "day",
   "year_hint": null,
   "basis": "GenMol NIM 2.0.0 release notes",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://docs.nvidia.com/nim/bionemo/genmol/2.0.0/release-notes.html",
   "verification": "partial",
   "description": "NIM 2.0.0 moves to the NV-GenMol-89M-v2 checkpoint; GitHub has had only maintenance commits since",
   "note": "Release notes say 2026-04-24 and the NGC container listing 2026-05-07; v2 weights had appeared in the repository earlier"
  },
  "status": "active",
  "status_note": "New version released within 12 months",
  "what_it_does": "Writes molecules as SAFE fragment sequences and generates them in parallel by masked discrete diffusion, with fragment remasking to explore chemical space; one pretrained model switches by prompting between de novo, fragment-constrained generation and lead optimisation. Unlike reinforcement-learning frameworks such as REINVENT 4 it needs no retraining per task; it belongs to the NVIDIA BioNeMo / Clara line.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "The SAFE dataset from ZINC and UniChem, about 946 million molecules",
    "source_url": "https://arxiv.org/abs/2501.06158"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2501.06158"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2501.06158"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-19-pmo",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2501.06158",
    "confidence": "high",
    "second_hand": false,
    "note": "Self-reported on the 23 PMO tasks under the original protocol, mean of 3 runs; baseline numbers are copied from other papers rather than rerun; the repository says v2 scores below v1"
   }
  ],
  "other_evaluations": [
   {
    "label": "De novo and fragment-constrained generation",
    "category": "own-task-set",
    "detail": "Validity, uniqueness and quality metrics, author-reported; most models are near saturation on these",
    "source_type": "self"
   },
   {
    "label": "Lead optimisation tasks",
    "category": "own-task-set",
    "detail": "Success on 30 lead-optimisation tasks, author-reported",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one discrete-diffusion model unifies several molecule-generation tasks, leads prior methods on PMO and ships open weights.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party rerun in a shared environment or wet-lab validation was found",
  "background": {
   "org_profile": "NVIDIA, founded in 1993 and based in Santa Clara, California, is a GPU and accelerated-computing company; Clara and BioNeMo are its open model and tool platforms for biology and drug discovery.",
   "funding": "A listed company whose drug-discovery work sits in its internal healthcare unit; its venture arm NVentures invests in drug-discovery companies such as Terray and Iambic.",
   "products_partnerships": "GenMol is offered as a NIM microservice on build.nvidia.com and NGC; KERMT, built with Merck, sits in the same BioNeMo line.",
   "related_models": [
    "model-reinvent-4",
    "model-saturn",
    "model-kermt"
   ]
  },
  "reading_notes": [
   "The PMO result is author-reported with baselines copied from other papers, so the margin should be discounted.",
   "v2 scores below v1 on PMO; cite the version.",
   "There is no wet lab, so synthesisability and activity of generated molecules are unknown.",
   "The strength is multitask unification and open weights rather than best-in-class on any one task."
  ],
  "caveats": [
   "Baselines were not rerun in the same environment.",
   "Two sources disagree on the latest version date (2026-04-24 vs 2026-05-07).",
   "No GuacaMol or MOSES results are reported."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2501.06158",
    "title": "arXiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/NVIDIA-Digital-Bio/genmol",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://docs.nvidia.com/nim/bionemo/genmol/2.0.0/release-notes.html",
    "title": "NIM 2.0.0 release notes",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://catalog.ngc.nvidia.com/orgs/nvidia/clara/resources/genmol_v1",
    "title": "NGC v1 resource page",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/resources/genmol_v2?version=1.0",
    "title": "NGC v2 resource page",
    "kind": "package-registry",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-01",
    "now": "2025-01-10 (arXiv v1 submission date)",
    "source_url": "https://arxiv.org/abs/2501.06158",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "NGC v2 weights updated 2025-10",
    "now": "2026-04-24 NIM 2.0.0 (the NGC container page says 2026-05-07)",
    "source_url": "https://docs.nvidia.com/nim/bionemo/genmol/2.0.0/release-notes.html",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "release_milestones.weights_release"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-genmol/",
   "en": "/bio/benchmark/models/genmol/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-germinal",
  "kind": "model",
  "short_slug": "germinal",
  "names": {
   "display": "Germinal",
   "full": null,
   "aliases": [
    "Germinal"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single pipeline with no family of versions",
  "domain": "antibody",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Open-source de novo antibody and nanobody design pipeline that redesigns only the CDRs",
  "org": {
   "names": [
    "Stanford University (chemical engineering)",
    "Arc Institute"
   ],
   "global_entity_ids": [
    "arc-institute"
   ]
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (antibody and nanobody CDR redesign)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code is open; it trains no weights of its own and depends on an antibody language model and a structure-modelling package whose licences are described second-hand"
  },
  "first_release": {
   "date": "2025-09-24",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1 posted (version API)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.19.677421",
   "verification": "verified",
   "note": "The site recorded 2025-09-19, which is the date inside the DOI string"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-09-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.19.677421",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-09-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.19.677421",
    "verification": "verified",
    "note": "v2 was posted 2025-09-25"
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-04-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Latest repository commit (no tagged release)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/SantiagoMille/germinal",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-09-24",
    "precision": "day",
    "year_hint": null,
    "basis": "News post from one of the participating institutes",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-04-24",
   "precision": "day",
   "year_hint": null,
   "basis": "Repository commits with no tagged release",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/SantiagoMille/germinal",
   "verification": "partial",
   "description": "Repository commits add scFv fixes, dependencies on an antibody language model and a scoring tool, and co-folding restraint fixes; there is no tagged release and no journal version was found.",
   "note": "The commits mix fixes and dependencies, so whether this is a substantive update is arguable"
  },
  "status": "active",
  "status_note": "Repository commits within 12 months but no tagged release",
  "what_it_does": "Redesigns the complementarity-determining regions on a fixed framework, scores binding with a co-folding model and constrains antibody-likeness with an antibody language model, aiming at epitope-directed antibodies without animal immunisation. It follows the same backpropagation-through-a-structure-model route as BindCraft, differing in its antibody focus and the added language model, and serves as the open academic counterpart to closed commercial models such as JAM-2 and Latent-X2.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "No dedicated training; it reuses off-the-shelf weights from a co-folding model and two antibody language models, built on public sequence and structure archives",
    "source_url": "https://doi.org/10.1101/2025.09.19.677421"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Wet lab is for validation only, with testing described second-hand as involving a gene-synthesis company",
    "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Four-target nanobody campaign",
    "targets": 4,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Four targets with dozens of designs each, first screened by a reporter assay then measured for binding; the exact numerator and denominator could not be verified this round",
    "success_rate": null,
    "affinity_best": null,
    "version": "bioRxiv v1/v2",
    "source_type": "self",
    "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
    "date": {
     "date": "2025-09-24",
     "precision": "day",
     "year_hint": null,
     "basis": "Institute news post",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
     "verification": "verified",
     "note": null
    },
    "verification": "unverified",
    "executor": "The preprint places the experiments in academic labs, with a gene-synthesis company involved in testing per second-hand sources",
    "unverified_marker": "Numerators, denominators, the success-rate range and affinity values could not be verified this round, so no numbers are given",
    "note": "The site previously published per-target hit counts, a 4-22% success range and nanomolar affinities; the full text was inaccessible this round"
   },
   {
    "campaign": "Epitope check",
    "targets": 1,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Alanine substitutions at predicted interface residues to see whether binding drops; no design count is given",
    "success_rate": null,
    "affinity_best": null,
    "version": "bioRxiv",
    "source_type": "self",
    "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
    "date": {
     "date": "2025-09-24",
     "precision": "day",
     "year_hint": null,
     "basis": "Institute news post",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
     "verification": "verified",
     "note": null
    },
    "verification": "unverified",
    "executor": "The same academic labs as the main campaign",
    "unverified_marker": "The size of the drop in binding could not be verified this round, so no number is given",
    "note": "The experiment speaks to where the designs bind, not to a success rate"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Four-target nanobody campaign",
    "category": "wet-lab-validation",
    "detail": "Dozens of designs on each of four targets, reporter screen then binding measurement; values unverifiable this round",
    "source_type": "self"
   },
   {
    "label": "Epitope check",
    "category": "wet-lab-validation",
    "detail": "Alanine substitutions at predicted interface residues to observe the change in binding",
    "source_type": "self"
   },
   {
    "label": "Internal ablations",
    "category": "own-task-set",
    "detail": "Compares co-folding guidance alone with added custom losses, and the antibody-language-model weight on developability and humanness measures",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: an open pipeline that, without animal immunisation, yields nanomolar binders on all four tested targets from under about a hundred designs each, with humanness improved by antibody-language-model guidance.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation or independent reproduction was found, and it has no pairing among the 44 benchmarks here",
  "background": {
   "org_profile": "A collaboration between a chemical engineering lab at a US university and a non-profit research institute in California, with no company entity; the institute's news post describes it as student-led.",
   "funding": "Funding is listed in the preprint acknowledgements with no amounts given.",
   "products_partnerships": "The output is the open pipeline itself; the same institute also produces genome and single-cell models such as Evo 2 and State.",
   "related_models": [
    "model-bindcraft",
    "model-jam-2",
    "model-latent-x2",
    "model-evo-2"
   ]
  },
  "reading_notes": [
   "Four targets with dozens of designs each is a small sample, and the work is an unreviewed preprint.",
   "The hit counts and success range previously published here could not be verified this round; the full text must be re-obtained before citing.",
   "The target list includes known easier cases, and generalisation to hard targets is unevidenced.",
   "Its value is as an open, inspectable counterpart to closed commercial antibody design models rather than scale."
  ],
  "caveats": [
   "All wet-lab figures were unverifiable this round, so the page gives no numbers.",
   "The first-release date is corrected away from the date inside the DOI string.",
   "The repository has no tagged release, so the latest update is a commit date.",
   "The pipeline trains no weights of its own and is bounded by its upstream models.",
   "It has no pairing among the 44 benchmarks covered here."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2025.09.19.677421",
    "title": "Preprint 2025",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.19.677421",
    "title": "Preprint version API record",
    "kind": "archive",
    "date": null
   },
   {
    "url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
    "title": "Institute news post",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/SantiagoMille/germinal",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-09-19 (date in the DOI string)",
    "now": "2025-09-24 (preprint v1 posted, version API)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.19.677421",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Per-target hit counts, 4–22% success rates and 140–560 nM affinities",
    "now": "The full text was rate-limited this round; all marked unverified and the numbers withdrawn",
    "source_url": "https://arcinstitute.org/news/building-germinal-for-ai-designed-antibody-molecules",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation",
   "release_milestones.code_release",
   "latest_update",
   "open_status.code_license",
   "open_status.weights",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": "arc-institute",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-germinal/",
   "en": "/bio/benchmark/models/germinal/"
  },
  "names_full_null_reason": "The paper uses a single word as the name with no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-gpt-rosalind",
  "kind": "model",
  "short_slug": "gpt-rosalind",
  "names": {
   "display": "GPT-Rosalind",
   "full": null,
   "aliases": [
    "GPT-Rosalind"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "The developer calls it the first model in a life-sciences series with no other members yet",
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "Life-sciences reasoning model offered through controlled access",
  "org": {
   "names": [
    "OpenAI"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Life-sciences reasoning LLM",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "No public weights and access is controlled; the companion life-sciences plugin sits in a public repository"
  },
  "first_release": {
   "date": "2026-04-16",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer launch announcement",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://openai.com/index/introducing-gpt-rosalind",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-04-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Launch announcement",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://openai.com/index/introducing-gpt-rosalind",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-04-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Launch announcement",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://openai.com/index/introducing-gpt-rosalind",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09-11",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer update announcement",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind",
   "verification": "partial",
   "description": "Exits research preview and opens globally to eligible organisations through controlled access, covering the enterprise assistant, the coding agent and an interface for internal tools.",
   "note": "The official page returns 403 to fetching and the event was confirmed through a search summary; a further related announcement's date was not retrieved"
  },
  "status": "active",
  "status_note": "Launch and a product update within 12 months",
  "what_it_does": "A reasoning model for biology, drug discovery and translational medicine, used to weigh evidence, propose hypotheses, design experiments and carry out multi-step research tasks, with a companion plugin reaching more than fifty public multi-omics databases, literature sources and biology tools at inference time. Unlike open agents such as Biomni and Kosmos it is a base model behind controlled access, and unlike frontier general models it has undergone life-sciences post-training by undisclosed means.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "The databases the plugin reaches are inference-time tools, not training data",
    "source_url": "https://openai.com/index/introducing-gpt-rosalind"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No public wet-lab validation; partners describe only applying or assessing it in their own workflows",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-18-labbench",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://openai.com/index/introducing-gpt-rosalind",
    "confidence": "medium",
    "second_hand": false,
    "note": "Covers 11 of the second version's 15 tasks, with only qualitative statements, no published scores and no entry on the official board"
   },
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://the-decoder.com/openai-launches-gpt-rosalind-a-reasoning-model-built-for-life-sciences-research/",
    "confidence": "medium",
    "second_hand": true,
    "note": "The official text claims only a lead among models with public scores and names no version; the figures appear solely in second-hand reports that disagree with each other"
   },
   {
    "benchmark_id": "bench-14-pubmedqa",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://openai.com/index/introducing-gpt-rosalind",
    "confidence": "high",
    "second_hand": true,
    "note": "Neither the official page nor reports mention it; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-23-bioasq",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://openai.com/index/introducing-gpt-rosalind",
    "confidence": "high",
    "second_hand": true,
    "note": "Neither the official page nor reports mention it; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-41-bioml-bench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://openai.com/index/introducing-gpt-rosalind",
    "confidence": "high",
    "second_hand": true,
    "note": "Neither the official page nor reports mention it; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-42-biomysterybench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://openai.com/index/introducing-gpt-rosalind",
    "confidence": "high",
    "second_hand": true,
    "note": "Neither the official page nor reports mention it; the cell rests on this site's own pages and must display as unconfirmed"
   }
  ],
  "other_evaluations": [
   {
    "label": "Developer's own life-sciences evaluation",
    "category": "own-task-set",
    "detail": "750 expert-written tasks with rubrics designed by outside experts and scored against them by another model from the same organisation",
    "source_type": "self"
   },
   {
    "label": "Internal capability evaluations",
    "category": "comparison-vs-baseline-only",
    "detail": "Compared with the organisation's earlier models across chemistry, protein understanding, phylogenetics, experiment design and tool use, reported only second-hand",
    "source_type": "self"
   },
   {
    "label": "Absent from the official boards",
    "category": "public-benchmark-not-on-site",
    "detail": "Neither of the two relevant official boards lists it, so the self-reported figures cannot be checked against them",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: the first life-sciences-specific reasoning model, leading among models with public scores on one bioinformatics analysis evaluation and beating their own previous model on part of a lab-protocol evaluation, with the clearest gain on molecular cloning questions.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Neither relevant official board lists it and no third-party evaluation was found",
  "background": {
   "org_profile": "The developer is an AI lab and company based in San Francisco; this is the first model in its life-sciences line.",
   "funding": "On 31 March 2026 it announced a $122B round at an $852B post-money valuation, led by three companies with existing shareholders continuing; the split is not officially disclosed.",
   "products_partnerships": "Launch partners include two biopharmaceutical companies, a non-profit institute and a laboratory instruments company, with another pharmaceutical company added in a June 2026 update; earlier collaborations involved two pharmaceutical companies and an AI drug development company.",
   "related_models": [
    "model-biomni",
    "model-kosmos",
    "model-co-scientist",
    "model-frontier-llms"
   ]
  },
  "reading_notes": [
   "Both evaluations were run by the developer with version, prompts and tool settings undisclosed, and neither official board lists it, so nothing can be compared with board scores.",
   "The official text is qualitative and the figures appear only in second-hand reports that contradict each other, so they must be treated as second-hand.",
   "It leads on only part of the protocol evaluation, which reads as strength in one kind of protocol design rather than an across-the-board lead.",
   "Controlled access makes outside reproduction hard, and the in-house evaluation is scored by another model from the same organisation, so judge and subject share a source."
  ],
  "caveats": [
   "Neither weights nor training data are public and there is no system card.",
   "The official page returns 403 to fetching, so several points could be checked only via cache or search summaries.",
   "Four benchmark pairings rest on this site's own pages and must display as unconfirmed.",
   "There is no public wet-lab validation; partners describe only use within their own workflows.",
   "The in-house evaluation's scorer and the model under test come from the same organisation."
  ],
  "sources": [
   {
    "url": "https://openai.com/index/introducing-gpt-rosalind",
    "title": "Developer launch announcement",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind",
    "title": "Developer update announcement",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://the-decoder.com/openai-launches-gpt-rosalind-a-reasoning-model-built-for-life-sciences-research/",
    "title": "Media report (second-hand)",
    "kind": "second-hand",
    "date": null
   },
   {
    "url": "https://help.openai.com/en/articles/20001193",
    "title": "Developer help page",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "latest_update",
   "training_data.a_public",
   "training_data.b_private_licensed",
   "training_data.c_own_wetlab_for_training",
   "benchmarks_reported",
   "other_evaluations"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-gpt-rosalind/",
   "en": "/bio/benchmark/models/gpt-rosalind/"
  },
  "names_full_null_reason": "A product name with no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1,
    "disputed-or-unconfirmed": 5
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-hcablm",
  "kind": "model",
  "short_slug": "hcablm",
  "names": {
   "display": "HCAbLM",
   "full": null,
   "aliases": [
    "HCAbLM"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single model; the group's separate generative HCAb model is a different model and is not merged",
  "domain": "antibody",
  "domains_secondary": [],
  "one_liner": "Nona Biosciences language model trained only on fully human heavy-chain-only antibodies, aimed at developability",
  "org": {
   "names": [
    "Nona Biosciences",
    "HBM Holdings (Harbour BioMed)"
   ],
   "global_entity_ids": []
  },
  "org_type": "pharma",
  "org_type_note": null,
  "country": [
   "CN",
   "US"
  ],
  "model_class": "Antibody LM (developability)",
  "open_status": {
   "code": "unclear",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The press release and preprint page do not say whether weights or code are public"
  },
  "first_release": {
   "date": "2026-09-15",
   "precision": "day",
   "year_hint": null,
   "basis": "bioRxiv preprint v1 posted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.09.10.750553",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-09-15",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.09.10.750553",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-09-15",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.09.10.750553",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-09-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Company press release",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.biospace.com/press-releases/nona-biosciences-successfully-develops-worlds-first-language-model-trained-on-fully-human-heavy-chain-only-antibodies",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09-15",
   "precision": "day",
   "year_hint": null,
   "basis": "Only the v1 preprint exists",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.09.10.750553",
   "verification": "verified",
   "description": "No later sourced event found (as of 2026-11-16); no new version, code or weights found",
   "note": null
  },
  "status": "active",
  "status_note": "First released 2026-09; no later version yet",
  "what_it_does": "A 366M-parameter language model that learns only the sequence grammar of fully human heavy-chain-only antibodies (HCAbs); the preprint says HCAbs differ from conventional human VH mainly in CDR1/2, CDR3 architecture and parts of the framework. Frozen representations are transferred to SEC purity, HIC behaviour and thermal stability. It is not a structure or design model, unlike IgGM or RFantibody; its comparators are general and antibody language models such as ESM3.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "Whether public data were mixed in is not disclosed",
    "source_url": "https://www.biospace.com/press-releases/nona-biosciences-successfully-develops-worlds-first-language-model-trained-on-fully-human-heavy-chain-only-antibodies"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://www.biospace.com/press-releases/nona-biosciences-successfully-develops-worlds-first-language-model-trained-on-fully-human-heavy-chain-only-antibodies"
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "Immune repertoires from the company's transgenic mouse platform: 73 immunised mice, 31.8M fully human HCAb sequences",
    "source_url": "https://www.biospace.com/press-releases/nona-biosciences-successfully-develops-worlds-first-language-model-trained-on-fully-human-heavy-chain-only-antibodies"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "SEC, HIC and thermal-stability measurements are used for transfer evaluation",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.09.10.750553v1"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Sequence distribution analysis",
    "category": "own-task-set",
    "detail": "Compares HCAb and conventional human VH sequences; shifts concentrate in CDRs and some framework regions",
    "source_type": "self"
   },
   {
    "label": "Frozen-embedding transfer to developability",
    "category": "own-task-set",
    "detail": "Predicts SEC purity, HIC behaviour and thermal stability with grouped internal validation and retrospective cross-project evaluation, all at the developer",
    "source_type": "self"
   },
   {
    "label": "Press-release baseline claim",
    "category": "comparison-vs-baseline-only",
    "detail": "Claims to beat IgLM, AbLang and a 6B-parameter ESM model on SEC and HIC; no numbers and the dataset is unnamed",
    "source_type": "self"
   },
   {
    "label": "FLAb and GDPa1",
    "category": "public-benchmark-not-on-site",
    "detail": "Not pairings: the FLAb2 model list does not include it, and the public evaluation in the press release is unnamed and cannot be taken as GDPa1",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a 366M model trained on 31.8M fully human HCAb sequences represents developability better than larger general or antibody language models.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Only the press release and preprint exist; no third-party evaluation found",
  "background": {
   "org_profile": "Nona Biosciences was set up in 2022-11 as a wholly owned subsidiary of HBM Holdings (Harbour BioMed, HKEX 02142), offering discovery and preclinical services up to IND; at launch the group said its transgenic mouse platform had been used by over 50 companies and institutions.",
   "funding": "No separate funding for Nona; the group reported about $158M total 2025 revenue (annual results release), and the 2025-03 AstraZeneca deal included a $105M equity investment.",
   "products_partnerships": "Products include H2L2 and HCAb transgenic mouse platforms and the Hu-mAtrIx AI discovery platform; in 2025 the group had deals with AstraZeneca, Otsuka, Pfizer, Bristol Myers Squibb and others; a separate generative HCAb model was announced in 2025-10.",
   "related_models": [
    "model-iggm",
    "model-rfantibody",
    "model-esm3"
   ]
  },
  "reading_notes": [
   "The only antibody language model on the site trained on a private transgenic-mouse repertoire; the data are distinctive but there is no results table to compare.",
   "The ESM comparator is named only in press-release wording; the actual model should be checked in the preprint.",
   "FLAb2 and GDPa1 are the public tests it most needs; worth following up.",
   "Weight availability is unstated, so it cannot be reproduced externally."
  ],
  "caveats": [
   "Comparison claims come from a press release with no numbers and an unnamed dataset.",
   "All developability evaluation was done inside the developer.",
   "None of the four added antibody benchmarks on the site has a public score for it."
  ],
  "sources": [
   {
    "url": "https://www.biorxiv.org/content/10.64898/2026.09.10.750553v1",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.biospace.com/press-releases/nona-biosciences-successfully-develops-worlds-first-language-model-trained-on-fully-human-heavy-chain-only-antibodies",
    "title": "Company press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.biospace.com/harbour-biomed-launches-nona-biosciences-ideas-to-ind-preclinical-solutions-business-to-accelerate-global-biotherapeutic-innovation",
    "title": "Nona launch press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://en.prnasia.com/releases/global/harbour-biomed-reports-full-year-2025-financial-results-sustained-global-collaboration-underpins-long-term-growth-527384.shtml",
    "title": "Group 2025 annual results release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.biospace.com/press-releases/harbour-biomed-enters-into-global-strategic-collaboration-with-astrazeneca-to-discover-and-develop-next-generation-therapeutic-antibodies",
    "title": "AstraZeneca collaboration press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12767642/",
    "title": "FLAb2 paper",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "open_status.code",
   "open_status.weights",
   "training_data.a_public"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-hcablm/",
   "en": "/bio/benchmark/models/hcablm/"
  },
  "names_full_null_reason": "Neither the preprint nor the press release gives an expanded name",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-helixfold-s1",
  "kind": "model",
  "short_slug": "helixfold-s1",
  "names": {
   "display": "HelixFold-S1",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "HelixFold-S1"
   ],
   "ambiguous_aliases": [
    "HelixFold"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "A contact-guided sampling method added on top of HelixFold3, not a new base model",
  "org": {
   "names": [
    "Baidu"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Contact-guided sampling method",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code and weights promised after acceptance but not seen released; an online service is available"
  },
  "first_release": {
   "date": "2025-07-12",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2507.09251 v1 (per the earlier record; only the v3 date was verified)",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2507.09251",
   "verification": "partial",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-07-12",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1 (date not re-verified)",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2507.09251",
    "verification": "partial",
    "note": null
   },
   "journal_online": {
    "date": "2026-07-02",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Nature Machine Intelligence online (Crossref)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s42256-026-01264-2",
    "verification": "partial",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-07-19",
   "precision": "day",
   "event_kind": "paper-revision",
   "basis": "arXiv v3 revision",
   "description": "arXiv v3 revision, after journal online publication on 2026-07-02; code and weights still not confirmed released",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2507.09251",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication and paper revision within 12 months",
  "what_it_does": "Not a new base model but a sampling method on top of HelixFold3: it first predicts inter-chain contacts, then samples likely interfaces, reaching comparable accuracy with far fewer samples.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB entries before 2021-09-30 plus structure-database distillation data; antibody data from SAbDab",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Self-built PDB interface sets",
    "category": "own-task-set",
    "detail": "Post-2021 PDB complexes: antibody-protein 221, protein-protein 198, ligand 238, RNA 177, DNA 254",
    "source_type": "self"
   },
   {
    "label": "Sampling efficiency",
    "category": "comparison-vs-baseline-only",
    "detail": "Compared with baselines on 74 antibody complexes at about one-tenth the sampling",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report contact-guided sampling gives higher interface accuracy at about one-tenth the sampling, with the largest gain on antibody-antigen interfaces",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "A biocomputing team at Baidu, built on its own deep-learning framework, internally funded; headcount not disclosed",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "The same team also built HelixFold3 and two design platforms; no HelixFold4 found",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "All evaluations are self-built PDB subsets outside the 44 benchmarks; on the zero-pairing list under this round's rules",
   "Code and weights are unreleased, so the sampling-efficiency claim cannot be reproduced externally"
  ],
  "caveats": [
   "Neither the arXiv v1 date nor the journal online date is fully verified",
   "It is a method, not a base model; not directly comparable with co-folding models"
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2507.09251",
    "title": "arXiv preprint 2507.09251",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s42256-026-01264-2",
    "title": "Journal article DOI 10.1038/s42256-026-01264-2",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.event_kind",
    "was": "Journal publication",
    "now": "Paper revision; the journal event is listed separately under 2026-07-02",
    "source_url": "https://api.crossref.org/works/10.1038/s42256-026-01264-2",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.journal_online"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-helixfold-s1/",
   "en": "/bio/benchmark/models/helixfold-s1/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-iggm",
  "kind": "model",
  "short_slug": "iggm",
  "names": {
   "display": "IgGM / IgGM-FM",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "IgGM",
    "IgGM-FM"
   ],
   "ambiguous_aliases": []
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "IgGM",
     "id_or_null": null,
     "first_release": {
      "date": "2024-09-22",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "bioRxiv v1",
      "timestamp_utc": null,
      "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.19.613838",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "IgGM-FM",
     "id_or_null": null,
     "first_release": {
      "date": "2025-09-16",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "Foundation-model version preprint",
      "timestamp_utc": null,
      "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.12.675771",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "Covers the original IgGM and its foundation-model version IgGM-FM; the wet-lab results come from the IgGM-FM preprint"
  },
  "family_null_reason": null,
  "domain": "antibody",
  "domains_secondary": [],
  "one_liner": "Generative model for antibodies and nanobodies covering structure prediction, de novo design, maturation and humanization",
  "org": {
   "names": [
    "Tencent AI for Life Sciences Lab"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Antibody generative model",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Code and weights on GitHub under the MIT licence"
  },
  "first_release": {
   "date": "2024-09-22",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 preprint (repository creation date not used; earlier than the conference paper)",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.19.613838",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-09-22",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.19.613838",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-09-16",
   "precision": "day",
   "event_kind": "new-version",
   "basis": "IgGM-FM preprint posted",
   "description": "The IgGM-FM foundation-model preprint was posted; later GitHub activity is small fixes (last 2026-05-28)",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.12.675771",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only small fixes after 2025-09",
  "what_it_does": "One model covers complex structure prediction, de novo design, affinity maturation, inverse design and humanization for antibodies and nanobodies; the foundation-model version broadens the task range.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public antibody complexes released before 2022-12-31: 6,448 heavy-light and 1,907 single-chain complexes in 2,436 clusters",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Affinity measurements on several antigens used for validation",
    "source_url": null
   }
  },
  "wetlab_validation": [
   {
    "campaign": "PD-L1 de novo antibodies",
    "modality": "antibody",
    "targets": null,
    "designs_tested": 60,
    "hits": 7,
    "denominator_basis": "10,000 candidates were generated per CDR-length combination (24 combinations), filtered by edit distance and structure/frequency, and 60 sequences were tested, with 7/60 binders",
    "success_rate": "7/60 binders",
    "affinity_best": "KD 0.084–2.89 nM / KD 0.084–2.89 nM",
    "version": "bioRxiv v1",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675771v1.full",
    "date": null,
    "verification": "verified",
    "note": "The 10,000 figure is per length combination, not the pool the 60 were drawn from; 6 of the 7 blocked ligand binding in ELISA; developer-run"
   },
   {
    "campaign": "PD-L1 antibody inhibition (unverified)",
    "modality": "antibody",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Unverified: the previously published lowest IC50 value cannot be found in the v1 full text, so no figure is given",
    "success_rate": null,
    "affinity_best": null,
    "version": "bioRxiv v1",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675771v1.full",
    "date": null,
    "verification": "unverified",
    "note": "The earlier published figure is withdrawn and treated as unverified"
   },
   {
    "campaign": "Affinity maturation and humanization",
    "modality": "antibody",
    "targets": null,
    "designs_tested": 20,
    "hits": 5,
    "denominator_basis": "20 candidates selected by generation frequency, yielding 5 humanized antibodies",
    "success_rate": null,
    "affinity_best": "KD from 52.02 nM to 9.753 nM (5.3x); humanized variants 0.139–0.486 nM",
    "version": "bioRxiv v1",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675771v1.full",
    "date": null,
    "verification": "verified",
    "note": "All figures verified; developer-run"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-36-aintibody",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": null,
    "source_url": "https://github.com/TencentAI4S/IgGM",
    "confidence": "low",
    "verification": "partial",
    "note": "The team states a top-three placement in the affinity-optimization track; the organizer's per-team ranking is not cited and confidence is low"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": "Developers report one model handles antibody structure prediction, de novo design, maturation and humanization, with nM-level PD-L1 binders",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evaluation found",
  "background": {
   "org_profile": "An in-house lab in Shenzhen funded by the company, with university and Chinese Academy of Sciences collaborators",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "The sister system tFold shares authors and comparison targets",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-tfold",
    "model-rfantibody",
    "model-origin-1"
   ]
  },
  "reading_notes": [
   "IgGM's 7/60 and nM affinities must never be attributed to tFold; the two were previously mixed up on the site",
   "The AIntibody placement is only self-stated and not confirmed by organizer material"
  ],
  "caveats": [
   "The old lowest-IC50 figure cannot be found in the full text and is withdrawn",
   "Generation counts are per length combination and must not be read as a screening pool"
  ],
  "sources": [
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.09.19.613838",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.09.12.675771",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/TencentAI4S/IgGM",
    "title": "GitHub repository TencentAI4S/IgGM",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://iclr.cc/virtual/2025/poster/27650",
    "title": "Official page iclr.cc",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675771.full.pdf",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "The 10,000 figure was the pool the 60 were drawn from",
    "now": "The 10,000 figure is per CDR-length combination (24 combinations)",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675771v1.full",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Lowest IC50 5.29 nM",
    "now": "The value is not in the full text; treated as unverified with no figure",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675771v1.full",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-iggm/",
   "en": "/bio/benchmark/models/iggm/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-intellifold-2",
  "kind": "model",
  "short_slug": "intellifold-2",
  "names": {
   "display": "IntelliFold-2",
   "full": null,
   "full_null_reason": "Official materials use IntelliFold-2 only; no expanded name",
   "aliases": [
    "IntelliFold-2",
    "IntelliFold-v2"
   ],
   "ambiguous_aliases": [
    "IntelliFold",
    "IntFold",
    "IntFOLD"
   ]
  },
  "family": null,
  "family_null_reason": "Covers IntelliFold-2 (Flash/standard/Pro) only; IntFold counted as predecessor",
  "domain": "structure",
  "domains_secondary": [
   "antibody"
  ],
  "one_liner": "Hong Kong co-folding model billed as an AF3 improvement; Pro tier is server-only",
  "org": {
   "names": [
    "IntelliGen AI"
   ],
   "global_entity_ids": []
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "HK"
  ],
  "model_class": "Co-folding (protein/nucleic acid/small molecule)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Flash and standard code and weights are Apache-2.0 and commercial-use; the Pro tier is server-only and not reproducible"
  },
  "first_release": {
   "date": "2026-02-07",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release v2.0.0 (the same month's bioRxiv report was not date-checked)",
   "timestamp_utc": null,
   "source_url": "https://github.com/IntelliGen-AI/IntelliFold/releases",
   "verification": "partial",
   "note": "The predecessor IntFold release of 2025-07 is not this record's first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-02-07",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub release v2.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/IntelliGen-AI/IntelliFold/releases",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2026-02",
    "precision": "month",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv technical report (release notes)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.64898/2026.02.09.704787",
    "verification": "unverified",
    "note": "Month only; the exact day is unverified"
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-02-07",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub release v2.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/IntelliGen-AI/IntelliFold/releases",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2026-02-07",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Hugging Face card published the Flash and standard weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/intelligenAI/intellifold",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-06-23",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release v2.0.4",
   "description": "v2.0.3 moved inference onto the AlphaFold 3 JAX engine (HF card updated 06-15) and v2.0.4 followed on 06-23; later commits are steering fixes and maintenance",
   "timestamp_utc": null,
   "source_url": "https://github.com/IntelliGen-AI/IntelliFold/releases",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Tagged releases, with commits to 2026-10 as maintenance",
  "what_it_does": "Predicts co-folded complexes of proteins, nucleic acids and small molecules, positioned as an improvement on the AlphaFold3 architecture. Flash and standard builds are public while the Pro tier ships only with the hosted service, combining open weights with a commercial offering. It sits in the same open AF3-reproduction line as Protenix and SeedFold.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Reprocessed PDB plus self-distillation data; details not reported",
    "source_url": "https://doi.org/10.64898/2026.02.09.704787"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "FoldBench antibody-antigen / protein-ligand",
    "source_url": "https://doi.org/10.64898/2026.02.09.704787",
    "confidence": "high",
    "verification": "partial",
    "note": "The release notes show results only as images; no citable table in the text"
   },
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "PoseBusters V2 (README, steering option)",
    "source_url": "https://github.com/IntelliGen-AI/IntelliFold",
    "confidence": "medium",
    "verification": "partial",
    "note": "The README gives approximate values only and states no sampling or pose-selection rule, so it is not comparable across papers"
   }
  ],
  "other_evaluations": [
   {
    "label": "Predecessor IntFold self-report",
    "category": "self-report",
    "detail": "The 2025-07 report self-reported protein-protein and antibody-antigen results on FoldBench; that is the predecessor's protocol and the two generations must not be mixed",
    "source_type": "self"
   },
   {
    "label": "Controllable adapters and ranking",
    "category": "capability-claim",
    "detail": "The predecessor paper also claims allosteric-state, constraint and affinity adapters plus a training-free similarity ranking (self-reported)",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report gains over AlphaFold3 on FoldBench antibody-antigen and protein-ligand tasks; the release notes give no citable table (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party rerun or board entry found",
  "background": {
   "org_profile": "IntelliGen AI is headquartered in Hong Kong and, per media reports, was co-founded in 2024-10 to compete with big-lab drug discovery; headcount is undisclosed (second-hand).",
   "funding": "Media report an angel round of tens of millions of dollars; neither amount nor investors are official (second-hand, unverified).",
   "funding_null_reason": null,
   "products_partnerships": "Flash and standard weights are public and the Pro tier is offered through the server; no pharma partnership appears in public material.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3",
    "model-protenix",
    "model-seedfold"
   ]
  },
  "reading_notes": [
   "The headline claim sits on FoldBench but has no citable table, so it stays a vendor claim.",
   "The PoseBusters line exists only as README approximations under a different protocol.",
   "IntFold and this generation use different protocols; do not read their results together."
  ],
  "caveats": [
   "The Pro tier is closed, so its best server-side results cannot be reproduced.",
   "No official Runs N' Poses report was found, so that pairing is treated as absent.",
   "Company and funding details come from media reports, not official confirmation."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.64898/2026.02.09.704787",
    "title": "IntelliFold-2 bioRxiv technical report",
    "kind": "preprint",
    "date": {
     "date": "2026-02",
     "precision": "month",
     "event_kind": "preprint-posted",
     "basis": "bioRxiv posting month",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.64898/2026.02.09.704787",
     "verification": "unverified",
     "note": null
    }
   },
   {
    "url": "https://github.com/IntelliGen-AI/IntelliFold/releases",
    "title": "IntelliFold GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/intelligenAI/intellifold",
    "title": "Hugging Face model card",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2507.02025",
    "title": "Predecessor IntFold technical report",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.scmp.com/tech/tech-trends/article/3317704/hong-kong-start-intelligen-ai-aims-challenge-google-deepmind-drug-discovery",
    "title": "Press report on the company",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-06-16 (v2.0.3)",
    "now": "2026-06-23 (v2.0.4 is the later tagged release)",
    "source_url": "https://github.com/IntelliGen-AI/IntelliFold/releases",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.paper_v1.date",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-intellifold-2/",
   "en": "/bio/benchmark/models/intellifold-2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-intern-s",
  "kind": "model",
  "short_slug": "intern-s",
  "names": {
   "display": "Intern-S",
   "full": null,
   "aliases": [
    "Intern-S1",
    "Intern-S1-Pro",
    "Intern-S2"
   ],
   "ambiguous_aliases": [
    "Intern"
   ]
  },
  "family": "Intern-S",
  "family_null_reason": null,
  "domain": "text-agents",
  "domains_secondary": [
   "small-molecule",
   "protein"
  ],
  "one_liner": "Open multimodal scientific foundation series layering scientific corpora and specialised tokenisers on general LLMs, including a trillion-parameter mixture-of-experts version",
  "org": {
   "names": [
    "Shanghai AI Laboratory"
   ],
   "global_entity_ids": []
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "General scientific multimodal LLM",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Weights of each generation are public on the model hub with quantised variants and a domestic mirror; the code repository carries the same permissive licence"
  },
  "first_release": {
   "date": "2025-08-21",
   "precision": "day",
   "year_hint": null,
   "basis": "Paper v1 of the first generation; a repository creation date cannot serve as first release under this round's rule",
   "event_kind": "preprint-posted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2508.15763",
   "verification": "partial",
   "note": "The site recorded 2025-07-24, a model hub repository creation date; a 2025-07 public launch has no verifiable page, so the paper v1 is used"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-08-21",
    "precision": "day",
    "year_hint": null,
    "basis": "First-generation paper v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2508.15763",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-08-21",
    "precision": "day",
    "year_hint": null,
    "basis": "First-generation paper v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2508.15763",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-08-21",
    "precision": "day",
    "year_hint": null,
    "basis": "Official code repository",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/InternLM/Intern-S1",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2026-09-13",
    "precision": "day",
    "year_hint": null,
    "basis": "First publication of the second generation's release weights, with a quantised variant two days earlier",
    "event_kind": "weights-release",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/collections/internlm/intern-s2",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-09-13",
   "precision": "day",
   "year_hint": null,
   "basis": "First publication of the second generation's release weights",
   "event_kind": "new-weights",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/collections/internlm/intern-s2",
   "verification": "verified",
   "description": "That date is the first publication of the weights; the repository's 2026-09-24 last-modified timestamp is only a card edit, not a weights event. Whether the same organisation's separate 2026-10 and 2026-11 model line belongs to this series could not be confirmed",
   "note": null
  },
  "status": "active",
  "status_note": "A new generation's weights were released within 12 months",
  "what_it_does": "Continues pretraining a general large model on scientific corpora with specialised tokenisers and a vision encoder, then post-trains with multi-task and agent reinforcement learning across molecules, reactions, materials and microscopy images; later versions scale the expert count to a trillion parameters and add a domain-specialisation module that leaves the backbone untouched. It divides work with the same organisation's SciReasoner, a smaller model dedicated to structured reasoning, and unlike biology-specific bases such as Evo 2 it does not pretrain on sequences from scratch.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Scientific documents used as page images for visual pretraining, interleaved image-text data and an image retrieval corpus; agent training uses public executable task sets whose task and environment counts the paper lists",
    "source_url": "https://arxiv.org/abs/2608.13505"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2608.13505"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://github.com/InternLM/Intern-S1"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Neither the three papers nor the official repository reports wet-lab validation",
    "source_url": "https://github.com/InternLM/Intern-S1"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-14-pubmedqa",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://github.com/InternLM/Intern-S1",
    "confidence": "high",
    "second_hand": false,
    "note": "Neither the official repository nor the readable text of the three papers reports a result on that evaluation"
   },
   {
    "benchmark_id": "bench-18-labbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://github.com/InternLM/Intern-S1",
    "confidence": "high",
    "second_hand": false,
    "note": "Neither the official repository nor the readable text of the three papers reports a result on that evaluation"
   },
   {
    "benchmark_id": "bench-23-bioasq",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://github.com/InternLM/Intern-S1",
    "confidence": "high",
    "second_hand": false,
    "note": "Neither the official repository nor the readable text of the three papers reports a result on that evaluation"
   },
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://github.com/InternLM/Intern-S1",
    "confidence": "high",
    "second_hand": false,
    "note": "Neither the official repository nor the readable text of the three papers reports a result on that evaluation"
   },
   {
    "benchmark_id": "bench-41-bioml-bench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://github.com/InternLM/Intern-S1",
    "confidence": "high",
    "second_hand": false,
    "note": "Neither the official repository nor the readable text of the three papers reports a result on that evaluation"
   },
   {
    "benchmark_id": "bench-42-biomysterybench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://github.com/InternLM/Intern-S1",
    "confidence": "high",
    "second_hand": false,
    "note": "Neither the official repository nor the readable text of the three papers reports a result on that evaluation"
   }
  ],
  "other_evaluations": [
   {
    "label": "Instruction-style sequence and molecule benchmarks",
    "category": "public-benchmark-not-on-site",
    "detail": "Tasks on two public instruction datasets compared with several frontier general models, all self-reported",
    "source_type": "self"
   },
   {
    "label": "Protein QA, microscopy and chemistry benchmarks",
    "category": "public-benchmark-not-on-site",
    "detail": "Several evaluations outside this site's 44 with mixed outcomes, some still topped by closed frontier models",
    "source_type": "self"
   },
   {
    "label": "An author-defined protein binder task set",
    "category": "own-task-set",
    "detail": "A task set defined in the second generation's preview paper, compared with frontier general models",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report open multimodal scientific models that beat frontier general models on scientific sequence instruction tasks.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No outside-lab reproduction was found, and the evaluations use the same organisation's toolchain",
  "background": {
   "org_profile": "The developer is a government-backed new-type research institution in Shanghai, reported as founded in July 2020; the series began with a vision model built with a company in 2021 and later extended to language and science. Scale and budget are not public.",
   "funding": "Government funding; no commercial investment.",
   "products_partnerships": "The same organisation also runs SciReasoner and general vision and language model lines; the second generation's preview does reinforcement learning against several open and closed agent harnesses.",
   "related_models": [
    "model-scireasoner",
    "model-frontier-llms",
    "model-evo-2"
   ]
  },
  "reading_notes": [
   "Self-reported biology results centre on instruction-style tasks where the general comparators score very low, so the gap mostly reflects task format and dedicated training.",
   "Closed frontier models still beat it on microscopy and molecular knowledge evaluations, so it should not be read as leading across the board.",
   "Evaluations use the same organisation's toolchain, which limits third-party reproduction.",
   "It has no countable pairing with the benchmarks on this site and cannot be used for ranking here."
  ],
  "caveats": [
   "First release does not use the model hub repository creation date; a 2025-07 launch has no verifiable page, so this record uses the paper v1 and marks it unverified.",
   "2026-09-24 is only a repository last-modified timestamp and must not be presented as a weights event.",
   "Parameter counts differ between the repository and the papers, and sizes of different versions must not be mixed.",
   "Whether the same organisation's other model line from late 2026 belongs to this series is unconfirmed and it is not merged into this record."
  ],
  "sources": [
   {
    "url": "https://github.com/InternLM/Intern-S1",
    "title": "Official code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2508.15763",
    "title": "First-generation paper",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2603.25040",
    "title": "Trillion-parameter version paper",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2608.13505",
    "title": "Second-generation preview paper",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/collections/internlm/intern-s2",
    "title": "Second-generation model collection",
    "kind": "model-hub",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-09-24",
    "now": "2026-09-13 (weights first public; 09-24 is only the repository's last-modified time)",
    "source_url": "https://huggingface.co/collections/internlm/intern-s2",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2025-07-24 (model-hub repository creation date)",
    "now": "2025-08-21 (first-generation paper v1), marked unverified",
    "source_url": "https://arxiv.org/abs/2508.15763",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "release_milestones.code_release",
   "training_data.b_private_licensed"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-intern-s/",
   "en": "/bio/benchmark/models/intern-s/"
  },
  "names_full_null_reason": "A series name with no unified expanded form given officially",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 6
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-isodde",
  "kind": "model",
  "short_slug": "isodde",
  "names": {
   "display": "IsoDDE",
   "full": "Isomorphic Labs Drug Design Engine",
   "aliases": [
    "IsoDDE",
    "Isomorphic Labs Drug Design Engine"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single internal engine; AlphaFold 3 is its predecessor with its own record",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule",
   "antibody"
  ],
  "one_liner": "Isomorphic Labs closed drug design engine extending AlphaFold 3 co-folding and affinity",
  "org": {
   "names": [
    "Isomorphic Labs"
   ],
   "global_entity_ids": [
    "isomorphic-labs"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Co-folding, affinity, cryptic pockets",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Closed internal engine; only the official article and technical report are public"
  },
  "first_release": {
   "date": "2026-02-10",
   "precision": "day",
   "year_hint": null,
   "basis": "Official article and Zenodo technical report released the same day",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-02-10",
    "precision": "day",
    "year_hint": null,
    "basis": "Official article",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-02-10",
    "precision": "day",
    "year_hint": null,
    "basis": "Zenodo technical report (not peer reviewed)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://zenodo.org/records/18606681",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-02-10",
    "precision": "day",
    "year_hint": null,
    "basis": "Official article",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-02-10",
   "precision": "day",
   "year_hint": null,
   "basis": "No technical update found after release",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier",
   "verification": "verified",
   "description": "No later sourced event found (as of 2026-11-16); the 2026-05 Series B is a corporate event, not a new technical release",
   "note": null
  },
  "status": "product-only",
  "status_note": "Internal and partner use only; no public code or weights",
  "what_it_does": "An internal drug design engine extending AlphaFold3: co-folding on cases with low similarity to training data, small-molecule binding-affinity prediction and cryptic-pocket identification. Like the open Boltz-2 it covers both structure and affinity, but it is closed and cannot be reproduced externally; it is the main reference point for open AF3 reproductions.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "The report only says the cut-off matches AF3 (PDB before 2021-09-30); no dataset list",
    "source_url": "https://zenodo.org/records/18606681"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://zenodo.org/records/18606681"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "The report lists no internal experimental data",
    "source_url": "https://zenodo.org/records/18606681"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Only retrospective cases on published structures, no new wet lab",
    "source_url": "https://zenodo.org/records/18606681"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://zenodo.org/records/18606681",
    "confidence": "medium",
    "second_hand": false,
    "note": "The technical report self-reports on Runs N' Poses by training-similarity bin, focusing on the hardest bin; vendor statement, no third-party reproduction"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "self",
    "evidence_url": "https://zenodo.org/records/18606681",
    "confidence": "medium",
    "second_hand": false,
    "note": "Post-hoc use of CASP16 blind affinity data in a comparison figure, chart only with no single value; not participation"
   }
  ],
  "other_evaluations": [
   {
    "label": "Novel antibody-antigen interface set",
    "category": "own-task-set",
    "detail": "Own low-similarity antibody interface set compared with AF3 and Boltz-2; vendor-reported",
    "source_type": "self"
   },
   {
    "label": "Affinity comparisons (FEP+, OpenFE, ChEMBL time split)",
    "category": "comparison-vs-baseline-only",
    "detail": "Bar charts by protein class, claiming to beat ML and even physics methods; no summary value in the text, and a second-hand figure is not found in official material",
    "source_type": "self"
   },
   {
    "label": "Retrospective cryptic-pocket case",
    "category": "case-study",
    "detail": "Retrospective reproduction of a published experimental finding (cereblon case), not new validation",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: clearly better co-folding than AF3 on low-similarity cases, affinity prediction beating all deep-learning methods, and cryptic-pocket identification.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Closed; no third-party reproduction or blind-test result found, and blind rounds such as CASP17 have not reported",
  "background": {
   "org_profile": "Isomorphic Labs was spun out of Alphabet in 2021, is based in London and is a sister organisation of Google DeepMind, positioned as an AI drug design company; AlphaFold 3 was released jointly by the two.",
   "funding": "A first external round of $600M (2025-03) and a $2.1B Series B (2026-05), both led by Thrive Capital with Alphabet and GV; valuation undisclosed (company announcements).",
   "products_partnerships": "Research deals with Eli Lilly and Novartis from 2024-01 (about $82.5M upfront combined, per trade press); the company says new funds go to expanding IsoDDE and moving candidates to the clinic.",
   "related_models": [
    "model-alphafold3",
    "model-boltz-2"
   ]
  },
  "reading_notes": [
   "All figures come from the vendor technical report; the official article has none, and there is no third-party reproduction.",
   "\"Beats AF3\" is a relative comparison on the hardest similarity bin, not overall Runs N' Poses performance.",
   "Affinity is the commercially key claim, yet public evidence is comparison charts only, with no citable summary value.",
   "Being closed, outside checks must wait for partner disclosures or blind-test results."
  ],
  "caveats": [
   "Vendor-reported, non-peer-reviewed technical report; cannot be reproduced externally.",
   "The CASP16 affinity use is retrospective, not a competition entry.",
   "Affinity figures circulating second-hand are not in official material and should not be cited.",
   "Training data are undisclosed, so generalisation and leakage cannot be assessed externally."
  ],
  "sources": [
   {
    "url": "https://zenodo.org/records/18606681",
    "title": "IsoDDE technical report (Zenodo)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier",
    "title": "Official launch article",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://isomorphiclabs.com/articles/isomorphic-labs-announces-series-b-investment-round",
    "title": "Series B announcement",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://isomorphiclabs.com/articles/isomorphic-labs-announces-600m-external-investment-round",
    "title": "First external funding announcement",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "training_data.a_public",
   "training_data.b_private_licensed",
   "training_data.c_own_wetlab_for_training",
   "benchmarks_reported[0]",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-isodde/",
   "en": "/bio/benchmark/models/isodde/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1,
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developer but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-jam-2",
  "kind": "model",
  "short_slug": "jam-2",
  "names": {
   "display": "JAM / JAM-2",
   "full": null,
   "aliases": [
    "JAM",
    "JAM-2"
   ],
   "ambiguous_aliases": [
    "JAM"
   ]
  },
  "family": "JAM",
  "family_null_reason": null,
  "domain": "antibody",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Closed de novo antibody and VHH design models aimed at hard multi-pass membrane targets such as GPCRs",
  "org": {
   "names": [
    "Nabla Bio"
   ],
   "global_entity_ids": [
    "nabla-bio"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (antibody / VHH)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "No public weights or code; the material is a company-hosted technical report, not peer reviewed"
  },
  "first_release": {
   "date": "2025-11",
   "precision": "month",
   "year_hint": null,
   "basis": "No later than that month: the technical report was covered by a third party; the primary page carries no date",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.startuphub.ai/ai-news/ai-research/2025/generative-antibody-design-cracks-undruggable-targets/",
   "verification": "partial",
   "note": "The previous generation's preprint is from 2025-05; the company page's entry date format is ambiguous and is not used"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-11",
    "precision": "month",
    "year_hint": null,
    "basis": "Third-party coverage of the technical report",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.startuphub.ai/ai-news/ai-research/2025/generative-antibody-design-cracks-undruggable-targets/",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-05",
    "precision": "month",
    "year_hint": null,
    "basis": "Preprint of the previous generation",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-06",
   "precision": "month",
   "year_hint": null,
   "basis": "Follow-on report on T-cell-engaging multispecific antibodies against peptide-MHC class I targets",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.truecoursebio.com/about/blog/cryo-em-validates-an-ai-designed-antibody-at-0-93-a-a-collaboration-with-nabla-bio",
   "verification": "unverified",
   "description": "The follow-on report extends to peptide-MHC class I targets and includes a dual-variant antibody, with cryo-EM confirmation by an imaging service provider; the report itself carries no version date and the month comes from a third-party post, so it is treated as unverified",
   "note": null
  },
  "status": "active",
  "status_note": "A follow-on report within 12 months, dated only to the month and unverified",
  "what_it_does": "Designs antibodies and VHHs de novo for multi-pass membrane targets such as GPCRs, ion channels and transporters; the newer generation extends output to drug-like VHH-Fc and full-length mAb formats, supports user-specified epitopes and steers at inference time from validated molecules. It sits with closed antibody design efforts such as Chai-2 and Latent-X/X2, opposite the academic open route of Germinal.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "The technical report does not state training data sources",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "Pretraining is undisclosed; inference-time steering prompted by validated molecules is not training",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "All reported results come from newly run wet-lab screens",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Target set 1 (16 unseen targets)",
    "targets": 16,
    "designs_tested": null,
    "hits": 16,
    "denominator_basis": "16 targets, 45 designs per modality per target, a binary binding readout; hits counted per target",
    "success_rate": "16/16 targets",
    "affinity_best": null,
    "version": "technical report",
    "source_type": "self",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "date": {
     "date": "2025-11",
     "precision": "month",
     "year_hint": null,
     "basis": "Technical report",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Run by the developer; the report contains no independent experimental arm",
    "unverified_marker": null,
    "note": "Mean per-target success is 39% and 18% for the two modalities; of the 16 targets, 2 reached picomolar and 7 single-digit nanomolar"
   },
   {
    "campaign": "Epitope-directed set (10 further targets)",
    "targets": 10,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "20 user-specified epitopes per target and about 600 designs per epitope; the report gives no per-epitope denominator",
    "success_rate": null,
    "affinity_best": null,
    "version": "technical report",
    "source_type": "self",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "date": {
     "date": "2025-11",
     "precision": "month",
     "year_hint": null,
     "basis": "Technical report",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Run by the developer",
    "unverified_marker": null,
    "note": "The report states that on half the targets 30-70% of epitopes yielded binders; a range without per-target denominators"
   },
   {
    "campaign": "GPCR binding in native cells",
    "targets": 2,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Two chemokine receptors tested in cells; the report gives no design-count denominator",
    "success_rate": "11.7% and 3.8% (per-target rates, denominators not given)",
    "affinity_best": null,
    "version": "technical report",
    "source_type": "self",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "date": {
     "date": "2025-11",
     "precision": "month",
     "year_hint": null,
     "basis": "Technical report",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Run by the developer",
    "unverified_marker": null,
    "note": "The report also claims that test-time compute scaling raises success substantially; self-reported"
   },
   {
    "campaign": "Experiment-guided steering (agonist antibodies)",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Second-hand source only, no denominator",
    "success_rate": null,
    "affinity_best": null,
    "version": "secondary coverage",
    "source_type": "self",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "date": {
     "date": "2025-11",
     "precision": "month",
     "year_hint": null,
     "basis": "Second-hand coverage contemporaneous with the report",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
     "verification": "unverified",
     "note": null
    },
    "verification": "unverified",
    "executor": "Executor undisclosed",
    "unverified_marker": "The agonist count previously cited on the site is second-hand and unverified; no figure is given",
    "note": "Generating like molecules prompted by one validated molecule is a case study, not a measured evaluation"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Comparison with another closed model's published numbers",
    "category": "comparison-vs-baseline-only",
    "detail": "Compares published binding outcomes on overlapping targets rather than re-running the other model",
    "source_type": "self"
   },
   {
    "label": "Own wet-lab evaluation on hard targets",
    "category": "wet-lab-validation",
    "detail": "The target-set, epitope-directed and native-cell experiments are all developer-defined and map to no benchmark on this site",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report fully computational design of drug-like VHH-Fc and full-length mAbs with binders on all 16 unseen targets, epitope-level control and activity against GPCRs in native cells.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "The material is a company-hosted technical report, not peer reviewed, and no third-party re-test was found",
  "background": {
   "org_profile": "The developer is headquartered in Massachusetts as an AI-native antibody design company spun out of a university lab; headcount of about twenty-odd comes from public data aggregators.",
   "funding": "An $11M seed in 2021-12 led by two venture firms and a $26M Series A in 2024-05 led by one firm. A 2026 Series B listed by a data aggregator has no company announcement or second source and is not used.",
   "products_partnerships": "It collaborates with three large pharmaceutical companies; in 2025-10 it signed a second multi-year deal with one of them with potential value above a billion dollars including milestones, upfront undisclosed. Competitors include the developers of Chai-2 and Latent-X/X2.",
   "related_models": [
    "model-chai-2",
    "model-latent-x2",
    "model-germinal"
   ]
  },
  "reading_notes": [
   "All results are self-reported, not peer reviewed, and there is no independent experimental arm.",
   "\"Binders on all 16 targets\" is a per-target statement; per-design hit rates are much lower and the two must not be swapped.",
   "The target set overlaps another closed model's study to enable comparison, and the comparison is between published results.",
   "Pretraining data is entirely undisclosed, so it cannot be filed as public-data-only."
  ],
  "caveats": [
   "Neither the first nor the follow-on report has a primary-page date, so dates are month-level only.",
   "The agonist count is second-hand and that figure is no longer given on the site.",
   "A second-hand claim that the previous generation got one hit in twenty thousand designs could not be found in the report.",
   "It reports no benchmark on this site, so comparison with other models is only indirect, through each party's own wet-lab protocol."
  ],
  "sources": [
   {
    "url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "title": "Technical report PDF",
    "kind": "technical-report",
    "date": null
   },
   {
    "url": "https://www.startuphub.ai/ai-news/ai-research/2025/generative-antibody-design-cracks-undruggable-targets/",
    "title": "Third-party coverage (2025-11-20)",
    "kind": "news",
    "date": "2025-11-20"
   },
   {
    "url": "https://www.truecoursebio.com/about/blog/cryo-em-validates-an-ai-designed-antibody-at-0-93-a-a-collaboration-with-nabla-bio",
    "title": "Imaging provider blog (relating to the follow-on report)",
    "kind": "blog",
    "date": null
   },
   {
    "url": "https://www.biospace.com/nabla-bio-closes-11m-seed-financing-led-by-khosla-ventures-and-zetta-venture-partners",
    "title": "Seed round coverage",
    "kind": "news",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "Cited the number of agonist antibodies (300+)",
    "now": "The number is second-hand and unverified; removed",
    "source_url": "https://nabla-public.s3.us-east-1.amazonaws.com/2025_Nabla_JAM2.pdf",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "Recorded only up to the technical report",
    "now": "Added the 2026-06 follow-up report on peptide-MHC class I, marked unverified (month-level)",
    "source_url": "https://www.truecoursebio.com/about/blog/cryo-em-validates-an-ai-designed-antibody-at-0-93-a-a-collaboration-with-nabla-bio",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "latest_update",
   "training_data.a_public",
   "wetlab_validation[3]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-jam-2/",
   "en": "/bio/benchmark/models/jam-2/"
  },
  "names_full_null_reason": "The technical report gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No model page in the site-wide entity library; only the developer organisation has one",
  "status_head": null
 },
 {
  "id": "model-kermt",
  "kind": "model",
  "short_slug": "kermt",
  "names": {
   "display": "KERMT",
   "full": "Kinetic GROVER Multi-Task",
   "aliases": [
    "KERMT",
    "Contrastive KERMT",
    "NV-KERMT-70M-v2"
   ],
   "ambiguous_aliases": []
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "KERMT v1",
     "id_or_null": "model-kermt",
     "first_release": {
      "date": "2025-10-14",
      "precision": "day",
      "year_hint": null,
      "basis": "First preprint",
      "event_kind": "paper-revision",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2510.12719v1",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "Contrastive KERMT v2.0",
     "id_or_null": null,
     "first_release": {
      "date": "2026-06-09",
      "precision": "day",
      "year_hint": null,
      "basis": "Second preprint and weight upload",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2606.11508",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "v1 and the contrastive v2.0 are two papers with separate weights; results must not be carried across"
  },
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "Graph-transformer chemical pretrained model with multitask fine-tuning and acceleration for in-house ADMET prediction",
  "org": {
   "names": [
    "Merck & Co. (modeling and informatics)",
    "NVIDIA (BioNeMo team)"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (graph-transformer pretraining + multitask fine-tuning)",
  "open_status": {
   "code": "open",
   "weights": "research-only",
   "weights_license": "NVIDIA Open Model License (v2.0 weights)",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code is open; v2.0 weights are offered on a model hosting platform under an open model licence, while the pharma company's own page states no licence"
  },
  "first_release": {
   "date": "2025-10-14",
   "precision": "day",
   "year_hint": null,
   "basis": "First preprint (arXiv 2510.12719)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2510.12719v1",
   "verification": "partial",
   "note": "The arXiv id and date are carried over from the site record and were not re-fetched this round"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-10-14",
    "precision": "day",
    "year_hint": null,
    "basis": "First preprint",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2510.12719v1",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-10-14",
    "precision": "day",
    "year_hint": null,
    "basis": "First preprint",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2510.12719v1",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-03-23",
    "precision": "day",
    "year_hint": null,
    "basis": "GitHub v1.0.1 (v1.0.0 earlier)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/NVIDIA-Digital-Bio/KERMT/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-06-09",
    "precision": "day",
    "year_hint": null,
    "basis": "v2 weights uploaded to the model hosting platform",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/nvidia/NV-KERMT-70M-v2",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-06-10",
   "precision": "day",
   "year_hint": null,
   "basis": "GitHub release v2.0.0 (contrastive version)",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://github.com/NVIDIA-Digital-Bio/KERMT/releases",
   "verification": "verified",
   "description": "Contrastive v2.0: code release v2.0.0, v2 weights uploaded 2026-06-09 with a matching preprint (2026-06-09); commits through 2026-09-15 are documentation and tooling maintenance",
   "note": null
  },
  "status": "active",
  "status_note": "New version, new weights and a new preprint within 12 months",
  "what_it_does": "Adapts a third-party graph-transformer chemical pretraining architecture - atom and bond message passing plus a transformer - pretrained on tens of millions of molecules from two public compound libraries, with the emphasis on multitask fine-tuning and hardware-side acceleration and memory optimisation. It is positioned as a production ADMET model inside a pharma company, compared against the multitask variant of Chemprop and sitting on the same pretrained-ADMET track as MiniMol and MolE.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "25 public ADMET endpoints with 114,112 compounds plus another public set of 6 endpoints and 3,521 compounds; pretraining on 11 million compounds from two public libraries",
    "source_url": "https://arxiv.org/html/2510.12719v1"
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "In-house pharma ADMET data: 30 endpoints, 800,733 compounds measured through 2024, not public",
    "source_url": "https://arxiv.org/html/2510.12719v1"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "The in-house data are the company's historical measurements, not generated for this model",
    "source_url": "https://arxiv.org/html/2510.12719v1"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "In-house pharma ADMET endpoint set (time split)",
    "category": "own-task-set",
    "detail": "A time-split evaluation over 30 endpoints and about 800k compounds against a multitask graph network and another pretrained model; the data are not public and cannot be reproduced externally",
    "source_type": "self"
   },
   {
    "label": "Own public-literature ADMET set and a third-party public set",
    "category": "own-task-set",
    "detail": "25 public endpoints plus a six-endpoint public set; in the same paper these two reverse the ordering seen on the in-house set, so the ranking depends on the setting",
    "source_type": "self"
   },
   {
    "label": "Similarity-binned generalisation analysis",
    "category": "comparison-vs-baseline-only",
    "detail": "Comparison by chemical-similarity bin, with the authors reporting an advantage over the multitask graph-network baseline in every bin; all self-reported",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: multitask fine-tuning clearly beats non-pretrained graph-network baselines, with the advantage growing at larger data sizes, plus an accelerated implementation.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No site-benchmark pairing and no third-party evaluation found; the key results rest on non-public in-house data",
  "background": {
   "org_profile": "A modelling and informatics department at a large US pharma company co-authoring with the biological-models team of a GPU company; compute and funding come from both companies' internal R&D budgets, and the model is also listed in the GPU company's open-model catalogue.",
   "funding": null,
   "products_partnerships": "The architecture comes from an open graph-transformer pretraining model published by a Chinese internet company's lab; the code is open under the GPU company's digital-biology GitHub organisation and the pharma company published a corporate story page (2026-03-19).",
   "related_models": [
    "model-chemprop-v2",
    "model-minimol",
    "model-mole",
    "model-genmol"
   ]
  },
  "reading_notes": [
   "The most informative results come from the in-house dataset, which is not public and cannot be reproduced or checked externally.",
   "There are no site-benchmark results, so it cannot be compared directly with ADMET-AI or MiniMol.",
   "The advantage over graph-network baselines holds only in the multitask setting on the in-house assays; the same paper's public and third-party sets reverse the order, and all of it is self-reported.",
   "The TDC pairing does not hold: that benchmark does not appear in the paper."
  ],
  "caveats": [
   "Zero pairings: no results on any site benchmark.",
   "The key comparison depends on non-public in-house pharma data.",
   "v1 and the contrastive v2.0 are different weights; results cannot be carried across.",
   "The first-release date is carried over from the site record and was not re-fetched this round."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2510.12719v1",
    "title": "First preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2510.12719v1",
    "title": "First preprint full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2606.11508",
    "title": "Contrastive-version preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/NVIDIA-Digital-Bio/KERMT/releases",
    "title": "Code release list",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/nvidia/NV-KERMT-70M-v2",
    "title": "v2 weights page",
    "kind": "package-registry",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "Candidate pairing TDC",
    "now": "Not a pairing: the benchmark does not appear in the paper; the actual benchmarks are an internal set and two public sets",
    "source_url": "https://arxiv.org/html/2510.12719v1",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2025-10 preprint",
    "now": "2026-06-10 contrastive-learning version v2.0 (code, weights and new preprint)",
    "source_url": "https://github.com/NVIDIA-Digital-Bio/KERMT/releases",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "release_milestones.first_public_evidence"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-kermt/",
   "en": "/bio/benchmark/models/kermt/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-kosmos",
  "kind": "model",
  "short_slug": "kosmos",
  "names": {
   "display": "Kosmos / Edison Analysis",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "Kosmos",
    "Edison Analysis"
   ],
   "ambiguous_aliases": [
    "Finch"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "Edison Scientific's \"AI scientist\" that loops literature search and data analysis into cited reports",
  "org": {
   "names": [
    "Edison Scientific",
    "FutureHouse"
   ],
   "global_entity_ids": [
    "futurehouse"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "AI-scientist agent",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Not open source; available through the Edison platform and API"
  },
  "first_release": {
   "date": "2025-11-05",
   "precision": "day",
   "event_kind": "official-announcement",
   "basis": "Edison Scientific announcement (with arXiv 2511.02824)",
   "timestamp_utc": null,
   "source_url": "https://edisonscientific.com/articles/announcing-kosmos",
   "verification": "partial",
   "note": "arXiv v1 may be 2025-11-04, earlier than this date; unverified"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-11-04",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1 submission date (recheck finding, unverified)",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2511.02824",
    "verification": "unverified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-11-05",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Edison Scientific announcement",
    "timestamp_utc": null,
    "source_url": "https://edisonscientific.com/articles/announcing-kosmos",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2025-11-05",
   "precision": "day",
   "event_kind": "none-since-release",
   "basis": "Initial announcement; later items are partnership news, not new versions",
   "description": "No later sourced event found (as of 2026-11-16)",
   "timestamp_utc": null,
   "source_url": "https://edisonscientific.com/articles/announcing-kosmos",
   "verification": "verified",
   "note": "2026 partnership news with Incyte and Pfizer is not a model update"
  },
  "status": "product-only",
  "status_note": "Available only as a platform and API",
  "what_it_does": "Kosmos is positioned as an \"AI scientist\" that cycles between literature search and data analysis to produce cited research reports. Edison Analysis (formerly Finch) is its data-analysis agent, offered as a separate product.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Base LLM with public literature and data; training data not disclosed",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "original BixBench",
    "source_url": "https://edisonscientific.com/news/introducing-edison-analysis",
    "confidence": "low",
    "verification": "partial",
    "note": "Developer-reported; the number belongs to Edison Analysis, whose identity vs Kosmos is unclear; same company made BixBench; not on the BixBench3 board"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": null,
  "independent_evidence_null_reason": "Only a re-run by competitor Phylo",
  "background": {
   "org_profile": "Edison Scientific is the for-profit spin-out of the nonprofit lab FutureHouse (founded 2023-09) and also makes LAB-Bench and BixBench",
   "funding": "Announced a $70M seed round on 2025-12-18, led by Triatomic Capital and Spark Capital",
   "funding_null_reason": null,
   "products_partnerships": "Partnerships announced with Incyte (2026-05-19) and Pfizer (2026-10-29); founding subscriptions ended, new plans pending",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-robin",
    "model-biomni"
   ]
  },
  "reading_notes": [
   "BixBench comes from the same company, so look for third-party re-runs",
   "The site BixBench cell is actually an Edison Analysis number; its identity vs Kosmos is still to be clarified"
  ],
  "caveats": [
   "First-release date may be one day after arXiv v1; not re-checked",
   "Closed source; cannot be reproduced externally"
  ],
  "sources": [
   {
    "url": "https://edisonscientific.com/articles/announcing-kosmos",
    "title": "Official page edisonscientific.com",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2511.02824",
    "title": "arXiv preprint 2511.02824",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://edisonscientific.com/news/introducing-edison-analysis",
    "title": "Official page edisonscientific.com",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-kosmos/",
   "en": "/bio/benchmark/models/kosmos/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-la-proteina",
  "kind": "model",
  "short_slug": "la-proteina",
  "names": {
   "display": "La-Proteina",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "La-Proteina"
   ],
   "ambiguous_aliases": [
    "Proteina"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "protein",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "All-atom protein generation with flow matching in a partially latent space, co-designable up to 800 residues",
  "org": {
   "names": [
    "NVIDIA"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "All-atom protein generation",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "NVIDIA Open Model License",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code Apache-2.0; weights under the NVIDIA Open Model License"
  },
  "first_release": {
   "date": "2025-07-13",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2507.09466 v1",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2507.09466",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-07-13",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2507.09466",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-09-09",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Code added on GitHub",
    "timestamp_utc": null,
    "source_url": "https://github.com/NVIDIA-BioNeMo/la-proteina",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-05-27",
   "precision": "day",
   "event_kind": "new-version",
   "basis": "arXiv v2 posted",
   "description": "arXiv v2 posted; the last code push is a small fix on 2025-09-23",
   "timestamp_utc": null,
   "source_url": "https://export.arxiv.org/api/query?id_list=2507.09466",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "New paper version within 12 months",
  "what_it_does": "Generates all-atom proteins with flow matching in a partially latent space: an explicit backbone plus per-residue latents carrying sequence and side-chain detail; co-designable proteins reach 800 residues.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Two public structure-database subsets (~344.5k and ~46.94M structures); PDB not used",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No wet lab in the paper; a downstream model did validation",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-40-motifbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": null,
    "source_url": "https://github.com/blt2114/MotifBench",
    "confidence": "high",
    "verification": "verified",
    "note": "The official board does list this model; the Chinese model page saying there is no entry is wrong and must be fixed"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "NVIDIA is a listed GPU and AI company; this work comes from its generative-AI research lab and is internally funded; accepted at ICLR 2026",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "It is the base for Proteina-Complexa, and Dyno Psi follows the same line",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-proteina-complexa",
    "model-dyno-psi",
    "model-rfdiffusion2-3"
   ]
  },
  "reading_notes": [
   "Its only pairing is an official board entry; the site model page previously said none, a published error",
   "The paper has no wet lab; wet-lab evidence comes from a downstream model"
  ],
  "caveats": [
   "Training uses predicted-structure subsets only, not the experimental structure database",
   "Its metrics are computational self-consistency, not wet-lab feasibility"
  ],
  "sources": [
   {
    "url": "https://export.arxiv.org/api/query?id_list=2507.09466",
    "title": "Official page export.arxiv.org",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://github.com/NVIDIA-BioNeMo/la-proteina",
    "title": "GitHub repository NVIDIA-BioNeMo/la-proteina",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2507.09466",
    "title": "arXiv preprint 2507.09466",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "The Chinese model page said no MotifBench entry was found",
    "now": "The official board lists it; relation is official participation",
    "source_url": "https://github.com/blt2114/MotifBench",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-la-proteina/",
   "en": "/bio/benchmark/models/la-proteina/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-latent-x2",
  "kind": "model",
  "short_slug": "latent-x2",
  "names": {
   "display": "Latent-X / Latent-X2",
   "full": null,
   "full_null_reason": "Two model generations in one record, so there is no single full name",
   "aliases": [
    "Latent-X",
    "Latent-X2"
   ],
   "ambiguous_aliases": [
    "Latent-Y"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Latent-X",
     "id_or_null": null,
     "first_release": {
      "date": "2025-07-22",
      "precision": "day",
      "event_kind": "official-announcement",
      "basis": "Latent-X company press release (the preprint, 2025-07-25, is listed separately)",
      "timestamp_utc": null,
      "source_url": "https://secure.businesswire.com/news/home/20250722244481/en/",
      "verification": "partial",
      "note": "Unverified: that date belongs to the predecessor Latent-X while the X2 press release is 2025-12-16; treated as a family entry"
     }
    },
    {
     "name": "Latent-X2",
     "id_or_null": null,
     "first_release": {
      "date": "2025-12-16",
      "precision": "day",
      "event_kind": "official-announcement",
      "basis": "Company press release (preprint 2025-12-23)",
      "timestamp_utc": null,
      "source_url": "https://www.businesswire.com/news/home/20251216040546/en/",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "Covers Latent-X and Latent-X2 only; the Latent-Y design agent built on top is a separate product and its campaign figures belong to the Latent-Y preprint."
  },
  "family_null_reason": null,
  "domain": "protein",
  "domains_secondary": [
   "antibody",
   "generation"
  ],
  "one_liner": "Closed platform models for binder, macrocycle and antibody design, testing only tens of designs per target",
  "org": {
   "names": [
    "Latent Labs"
   ],
   "global_entity_ids": [
    "latent-labs"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Design (binder / antibody / macrocycle)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Weights are closed and access runs through a no-code platform; X2 requires an application or a commercial agreement"
  },
  "first_release": {
   "date": "2025-07-22",
   "precision": "day",
   "event_kind": "official-announcement",
   "basis": "Latent-X company press release (the preprint, 2025-07-25, is listed separately)",
   "timestamp_utc": null,
   "source_url": "https://secure.businesswire.com/news/home/20250722244481/en/",
   "verification": "partial",
   "note": "Unverified: that date belongs to the predecessor Latent-X while the X2 press release is 2025-12-16; treated as a family entry"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-07-22",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Latent-X company press release (the preprint, 2025-07-25, is listed separately)",
    "timestamp_utc": null,
    "source_url": "https://secure.businesswire.com/news/home/20250722244481/en/",
    "verification": "partial",
    "note": "Unverified: that date belongs to the predecessor Latent-X while the X2 press release is 2025-12-16; treated as a family entry"
   },
   "paper_v1": {
    "date": "2025-07-25",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Latent-X preprint (the X2 preprint is 2025-12-23)",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2507.19375",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-12-16",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Latent-X2 company press release",
    "timestamp_utc": null,
    "source_url": "https://www.businesswire.com/news/home/20251216040546/en/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-07-15",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Company press release: Latent-Y opened to researchers worldwide",
   "description": "Latent-Y, the antibody design agent built on X2 (launched 2026-03-23), opened to researchers worldwide with free daily allowances; no new X2 version or weights were found.",
   "timestamp_utc": null,
   "source_url": "https://secure.businesswire.com/news/home/20260715684640/en/",
   "verification": "partial",
   "note": "A product event, not a model version update to X2"
  },
  "status": "active",
  "status_note": "Product release events within 12 months, though no model version update",
  "what_it_does": "Latent-X targets mini-binders and macrocyclic peptides; X2 extends to drug-like antibodies (VHH, scFv) and reports unoptimised developability measures plus one ex vivo immunogenicity assessment. It shares the closed platform route with Chai-2 and JAM/JAM-2; unlike open models such as BoltzGen, its figures cannot be reproduced externally.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "Training data sources and scale are not disclosed",
    "source_url": "https://arxiv.org/abs/2512.20263"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "Whether private or licensed data is used is not disclosed",
    "source_url": "https://arxiv.org/abs/2512.20263"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "Not disclosed; must not be written up as 'public data only'",
    "source_url": "https://arxiv.org/abs/2512.20263"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Yes: the wet lab is used for validation and no feedback into training is claimed",
    "source_url": "https://arxiv.org/abs/2512.20263"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Latent-X macrocycle and mini-binder campaign",
    "modality": "macrocycle; mini-binder",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "A self-built target set with 30-100 designs per target; the source gives only an aggregate hit rate with no per-target numerators or denominators, so no rate is quoted",
    "success_rate": null,
    "affinity_best": "low nanomolar to picomolar (mini-binders, self-reported)",
    "version": "Latent-X (predecessor)",
    "source_type": "self",
    "source_url": "https://arxiv.org/abs/2507.19375",
    "date": {
     "date": "2025-07-25",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "Latent-X preprint",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2507.19375",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "Belongs to the predecessor model, and the head-to-head comparisons were run by the developer, not independently"
   },
   {
    "campaign": "Latent-X2 VHH / scFv campaign",
    "modality": "VHH; scFv",
    "targets": 18,
    "designs_tested": null,
    "hits": 9,
    "denominator_basis": "Eighteen targets with only 4-24 designs tested per target per modality",
    "success_rate": "binders on 9 of 18 targets (target-level rate)",
    "affinity_best": "picomolar to nanomolar (self-reported)",
    "version": "Latent-X2",
    "source_type": "self",
    "source_url": "https://arxiv.org/abs/2512.20263",
    "date": {
     "date": "2025-12-23",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "Latent-X2 preprint",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2512.20263",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "The source says VHH and scFv and never 'nanobody'; self-reported with no independent replication"
   },
   {
    "campaign": "Latent-Y agent campaigns",
    "modality": "nanobody",
    "targets": 9,
    "designs_tested": null,
    "hits": 6,
    "denominator_basis": "Nine targets with candidates generated by X2 and no human filtering; per-target hit rates run about 1%-28% of tested sequences",
    "success_rate": "binders on 6 of 9 targets (target-level rate)",
    "affinity_best": "single-digit nanomolar (self-reported)",
    "version": "Latent-Y (separate preprint)",
    "source_type": "self",
    "source_url": "https://arxiv.org/abs/2603.29727",
    "date": {
     "date": "2026-03-23",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Latent-Y launch (its preprint is separate)",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2603.29727",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "Source attribution corrected: these figures come from the Latent-Y preprint, not the 2025-12 press release and not the Latent-X2 paper"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Latent-X mini-binder / macrocycle campaign",
    "category": "wet-lab-validation",
    "detail": "A self-built target set with 30-100 designs per target, compared against other design models under matched conditions (self-reported)",
    "source_type": "self"
   },
   {
    "label": "Latent-X2 VHH / scFv campaign",
    "category": "wet-lab-validation",
    "detail": "Eighteen targets with 4-24 designs per target per modality and binders on nine; macrocycle binders were also made for one hard target",
    "source_type": "self"
   },
   {
    "label": "Developability panel",
    "category": "own-task-set",
    "detail": "Expression, aggregation, polyreactivity, hydrophobicity and thermal stability versus approved antibodies, with no optimisation or selection",
    "source_type": "self"
   },
   {
    "label": "Immunogenicity assay",
    "category": "case-study",
    "detail": "T-cell proliferation and cytokine release on de novo VHH binders for one target in a human donor panel",
    "source_type": "self"
   },
   {
    "label": "Latent-Y agent campaigns",
    "category": "user-or-partner-report",
    "detail": "Target-level success across nine nanobody campaigns, sourced to the Latent-Y preprint rather than the X2 paper",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report zero-shot VHH and scFv designs reaching picomolar to nanomolar affinity from tens of designs per target, with developability close to approved antibodies and low ex vivo immunogenicity (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Weights are closed and no third-party evaluation or independent replication was found",
  "background": {
   "org_profile": "The company was founded in 2023, headquartered in London with a San Francisco team, and positions itself as an AI-native protein design company; headcount is about 20 on aggregator data (2026-03, profile counts only).",
   "funding": "It emerged from stealth in 2025-02 with $50M, including a $40M Series A co-led by two funds with several other investors; no later round has been announced.",
   "funding_null_reason": null,
   "products_partnerships": "The products are Latent-X, Latent-X2 and the Latent-Y design agent launched 2026-03 and opened to researchers worldwide in 2026-07; no pharma deal values are public.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-chai-2",
    "model-jam-2",
    "model-alphaproteo",
    "model-boltzgen"
   ]
  },
  "reading_notes": [
   "The predecessor's comparison with AlphaProteo is self-reported on company-chosen targets and is not better on every target.",
   "Binders on 9 of 18 targets is a target-level rate, a different measure from per-target hit rates.",
   "Neither the training data nor the developability data can be checked externally.",
   "Both dates hold but mean different things: the company press release is 2025-12-16 and the preprint 2025-12-23."
  ],
  "caveats": [
   "Zero pairings: no site benchmark score and no competition entry record.",
   "One record covers two generations; the first-release date belongs to the predecessor Latent-X and is marked unverified.",
   "The Latent-Y campaign figures belong to a separate preprint and must not be counted as X2 results.",
   "Only 4-24 designs are tested per target, so target-level success is highly sensitive to the design budget."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2512.20263",
    "title": "Latent-X2 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2507.19375",
    "title": "Latent-X preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20251216040546/en/",
    "title": "Latent-X2 company press release",
    "kind": "press",
    "date": {
     "date": "2025-12-16",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Press release issued",
     "timestamp_utc": null,
     "source_url": "https://www.businesswire.com/news/home/20251216040546/en/",
     "verification": "partial",
     "note": null
    }
   },
   {
    "url": "https://secure.businesswire.com/news/home/20250722244481/en/",
    "title": "Latent-X company press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2603.29727",
    "title": "Latent-Y preprint (source of the campaign figures)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://secure.businesswire.com/news/home/20260715684640/en/",
    "title": "Latent-Y availability announcement",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "The nine campaigns and 67% target success were attributed to the 2025-12 press release or the Latent-X2 paper",
    "now": "Attribution corrected to the Latent-Y preprint and written as 6 of 9 targets",
    "source_url": "https://arxiv.org/abs/2603.29727",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "The first-release date was taken as Latent-X2's own release date",
    "now": "That date belongs to the predecessor Latent-X; the record becomes a family entry with an explicit boundary (unverified)",
    "source_url": "https://secure.businesswire.com/news/home/20250722244481/en/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "latest_update",
   "training_data.a_public",
   "training_data.b_private_licensed",
   "training_data.c_own_wetlab_for_training"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-latent-x2/",
   "en": "/bio/benchmark/models/latent-x2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-lingshu-cell",
  "kind": "model",
  "short_slug": "lingshu-cell",
  "names": {
   "display": "Lingshu-Cell",
   "full": null,
   "aliases": [
    "Lingshu-Cell",
    "灵枢·细胞"
   ],
   "ambiguous_aliases": [
    "Lingshu"
   ]
  },
  "family": null,
  "family_null_reason": "A single model with no later version",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Alibaba DAMO Academy generative cell world model using masked discrete diffusion over transcriptomes",
  "org": {
   "names": [
    "Alibaba DAMO Academy"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Single-cell generative / virtual cell",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Neither the project page nor the paper gives a code repository or weights; parameter count not reported"
  },
  "first_release": {
   "date": "2026-03-26",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv preprint v1",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://export.arxiv.org/api/query?id_list=2603.25240",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-03-26",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://export.arxiv.org/api/query?id_list=2603.25240",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-03-26",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://export.arxiv.org/api/query?id_list=2603.25240",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-03-26",
   "precision": "day",
   "year_hint": null,
   "basis": "Only arXiv v1 exists",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://export.arxiv.org/api/query?id_list=2603.25240",
   "verification": "verified",
   "description": "No later sourced event found (as of 2026-11-16); no new version, code or weights",
   "note": null
  },
  "status": "active",
  "status_note": "First released 2026-03, within 12 months",
  "what_it_does": "Discretises expression into tokens and learns the distribution of cell states over about 18k genes by masked discrete diffusion, generating unconditionally or conditioned on cell type, donor and perturbation; gene tokens are grouped and compressed to cut attention cost. Unlike STATE, which regresses pre- to post-perturbation states, it targets distributions rather than mean responses, and unlike CellFM its main output is not a cell embedding.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PARSE 10M PBMC, CZ CELLxGENE, scBaseCount, VCC H1 training data and public perturbation datasets",
    "source_url": "https://arxiv.org/pdf/2603.25240"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "Not disclosed in the paper; no use found",
    "source_url": "https://arxiv.org/pdf/2603.25240"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "Perturbation data were produced by third parties such as Arc and Parse",
    "source_url": "https://arxiv.org/pdf/2603.25240"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "retrospective-dataset-use",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/pdf/2603.25240",
    "confidence": "high",
    "second_hand": false,
    "note": "The authors evaluated post hoc on the public VCC H1 leaderboard data; the appendix labels it post-hoc and separates it from entries. Not a competition submission and not a board first place"
   }
  ],
  "other_evaluations": [
   {
    "label": "Cytokine perturbation prediction (PARSE PBMC)",
    "category": "own-task-set",
    "detail": "12 donors × 90 cytokine conditions against PertMean, STATE, scGPT and scVI; bar charts only, no per-metric table",
    "source_type": "self"
   },
   {
    "label": "Transcriptome distribution reproduction",
    "category": "own-task-set",
    "detail": "Reproduces distributions, marker genes and subtype proportions across nine tissues and several species; author-reported",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: generative modelling of cell-state distributions, the best mean rank in a post-hoc comparison on the VCC H1 genetic-perturbation board, and the best average on cytokine perturbation prediction.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Code and weights are not public; no third-party reproduction found",
  "background": {
   "org_profile": "DAMO Academy is Alibaba's research arm, announced in 2017-10 with a plan for RMB 100 billion over three years and seven labs in several countries; a big-tech lab, not separately funded.",
   "funding": "Funded from Alibaba's internal R&D budget; the team's own budget is not public.",
   "products_partnerships": "DAMO has other recent science projects such as a materials-discovery agent; the team discloses no pharma partnerships and relies on Arc Institute's VCC data.",
   "related_models": [
    "model-state",
    "model-cellfm",
    "model-stack",
    "model-tahoe-x1",
    "model-xtrimoscperturb-scfoundation"
   ]
  },
  "reading_notes": [
   "The best mean rank is driven by a few metrics while another sits near the bottom, so metric choice changes the conclusion.",
   "Evaluation was post hoc on public data and did not go through the blind challenge process.",
   "Independent VCC-era evaluations generally find simple baselines hard to beat, so its edge needs third-party reproduction.",
   "Without code or weights it cannot be cross-checked on other benchmarks."
  ],
  "caveats": [
   "The VCC relation is post-hoc data reuse, not counted as participation.",
   "Parameter count and licence are not published.",
   "All results are author-reported."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2603.25240",
    "title": "arXiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://alibaba-damo-academy.github.io/lingshu-cell-homepage",
    "title": "Project page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://virtualcellchallenge.org/leaderboard",
    "title": "VCC leaderboard",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://export.arxiv.org/api/query?id_list=2603.25240",
    "title": "arXiv version record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.cnbc.com/2017/10/11/alibaba-says-will-pour-15-billion-into-global-research-program.html",
    "title": "DAMO Academy launch coverage",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-lingshu-cell/",
   "en": "/bio/benchmark/models/lingshu-cell/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-lucaone",
  "kind": "model",
  "short_slug": "lucaone",
  "names": {
   "display": "LucaOne",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "LucaOne"
   ],
   "ambiguous_aliases": [
    "LucaCell",
    "LucaVirus"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "protein",
  "domains_secondary": [
   "genomics-rna"
  ],
  "one_liner": "Unified pretrained foundation model for DNA, RNA and protein sequences, about 1.8B parameters",
  "org": {
   "names": [
    "Alibaba Cloud",
    "Sun Yat-sen University"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": "Joint work by a cloud business unit and a university",
  "country": [
   "CN"
  ],
  "model_class": "Unified nucleic-acid and protein model",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code and weights Apache-2.0; the pretraining corpus is archived in a public genome database"
  },
  "first_release": {
   "date": "2024-05-14",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 preprint (repository creation date not used)",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.05.10.592927",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-05-14",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.05.10.592927",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-06-18",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Journal online publication (per the earlier record; Crossref fetch failed)",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s42256-025-01044-4",
    "verification": "partial",
    "note": null
   },
   "code_release": null,
   "weights_release": {
    "date": "2025-08-12",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "LucaOne weights uploaded to Hugging Face",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/api/models?author=LucaGroup",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-09-08",
   "precision": "day",
   "event_kind": "successor-model",
   "basis": "Successor LucaCell weights added on Hugging Face",
   "description": "Successor LucaCell weights released; LucaVirus came earlier in 2025 and the last GitHub commit is bug fixes",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/api/models?author=LucaGroup",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Successor weights released within 12 months",
  "what_it_does": "Unified semi-supervised multi-task pretraining over DNA, RNA and protein sequences at about 1.8B parameters, with a corpus spanning 169,861 species; downstream tasks use simple heads.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "DNA, RNA and protein sequences from 169,861 species archived in a public genome database; total counts not confirmed in the paper",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Eight downstream validation tasks",
    "category": "own-task-set",
    "detail": "Author-chosen DNA, RNA, protein and mixed-input tasks compared with several reference models",
    "source_type": "self"
   },
   {
    "label": "Central-dogma few-shot analysis",
    "category": "case-study",
    "detail": "Shows the DNA-protein correspondence the model learned in few-shot settings",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report a unified nucleic-acid and protein foundation model over 169,861 species that captures central-dogma relations and is competitive on eight downstream tasks",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Two third-party evaluations (2026) split: strong under supervised random splits but clearly weak zero-shot; viral-taxonomy results vary by split, and one listing marks its numbers as self-reported",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Published jointly by Alibaba Cloud (Hangzhou) and Sun Yat-sen University (Shenzhen); no separate funding",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "The same team also built a virus-discovery model and the successors LucaVirus and LucaCell",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-evo-2",
    "model-ntv3",
    "model-xtrimopglm"
   ]
  },
  "reading_notes": [
   "All self-evaluations are bespoke and outside the 44 benchmarks; on the zero-pairing list under this round's rules",
   "Third-party work shows a large gap between supervised and zero-shot settings; do not cite only one side"
  ],
  "caveats": [
   "The journal online date comes from the earlier record and was not re-checked via Crossref",
   "Total corpus counts are not confirmed in the paper"
  ],
  "sources": [
   {
    "url": "https://huggingface.co/api/models?author=LucaGroup",
    "title": "Hugging Face page api/models?author=LucaGroup",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.05.10.592927",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/LucaOne/LucaOne",
    "title": "GitHub repository LucaOne/LucaOne",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s42256-025-01044-4",
    "title": "Journal article DOI 10.1038/s42256-025-01044-4",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2511.02888",
    "title": "Third-party nucleic-acid model benchmark",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2605.25388",
    "title": "Third-party viral genomics benchmark",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "release_milestones.journal_online"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-lucaone/",
   "en": "/bio/benchmark/models/lucaone/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-massivefold",
  "kind": "model",
  "short_slug": "massivefold",
  "names": {
   "display": "MassiveFold",
   "full": null,
   "full_null_reason": "Official materials use MassiveFold only; no expanded name",
   "aliases": [
    "MassiveFold",
    "AFmassive"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single tool; its two inference back ends are not separate models",
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "Large-scale sampling and parallelisation around AF2, not a new model; hard interfaces gain most",
  "org": {
   "names": [
    "Lille University UGSF",
    "CNRS"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "FR"
  ],
  "model_class": "Sampling and parallelisation framework",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code is open source, inference uses the official AF2 weights, and the companion dataset ships under an open licence"
  },
  "first_release": {
   "date": "2024-04-30",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "Preprint of the tool paper (journal version 2024-11-11)",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s43588-024-00714-4",
   "verification": "verified",
   "note": "The site had treated 2025-05 as the first release; that is the CASP16-CAPRI data paper"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-04-30",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint of the tool paper",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s43588-024-00714-4",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-04-30",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Tool paper preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s43588-024-00714-4",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2024-11-11",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Tool paper journal publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s43588-024-00714-4",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": null,
    "precision": null,
    "event_kind": "code-release",
    "basis": "No sourced date for the first tagged release",
    "timestamp_utc": null,
    "source_url": "https://github.com/GBLille/MassiveFold/releases",
    "verification": "unverified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-11-12",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release v1.8.5",
   "description": "v1.8.5 tracks the new inference back end and refreshes install files and docs; the v1.8.x series began 2026-05, and the CASP16-CAPRI data paper appeared 2025-08-28",
   "timestamp_utc": null,
   "source_url": "https://github.com/GBLille/MassiveFold/releases",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Still releasing as of 2026-11",
  "what_it_does": "Runs AF2-Multimer repeatedly with many parameter sets, generating up to 8,040 structures per target and then selecting a few by confidence. It is a sampling and parallelisation tool rather than a new model, so it cannot be ranked against co-folding models.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Uses the official AF2 weights (PDB, UniRef and others); no new training",
    "source_url": "https://doi.org/10.1038/s43588-024-00714-4"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-22-capri",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "CASP16-CAPRI Round 57 (Round 58 data also released)",
    "source_url": "https://doi.org/10.1101/2025.05.26.653955",
    "confidence": "medium",
    "verification": "verified",
    "note": "Participation holds, but neither the paper nor the presentation reports an official placing"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "CASP16, five models per target selected by confidence",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified), and no official placing is reported"
   }
  ],
  "other_evaluations": [
   {
    "label": "Sampling-volume ablation",
    "category": "self-report",
    "detail": "The paper self-reports that cutting the per-target prediction count sharply costs little accuracy, and that difficulty can be predicted from ordinary confidence, so the tool can be reserved for hard cases (no figures)",
    "source_type": "self"
   },
   {
    "label": "Failure of the selection step",
    "category": "self-report",
    "detail": "The paper concludes good models are present in the sampled set but confidence cannot rank them to the top, which is the output bottleneck",
    "source_type": "self"
   },
   {
    "label": "Compute footprint",
    "category": "resource-use",
    "detail": "The project began at a 2023 hackathon and its large-scale sampling used a 300,000 GPU-hour national supercomputer allocation (per the paper)",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that large-scale sampling helps hardest interfaces most and that difficulty is predictable from ordinary confidence, so it can be used selectively (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No official placing is reported and no third-party rerun was found",
  "background": {
   "org_profile": "An academic team with no company: the UGSF laboratory of Lille University and CNRS (GBLille team), not Strasbourg.",
   "funding": "No funding amount is public; its CASP16 sampling used a 300,000 GPU-hour national supercomputer allocation (per the paper).",
   "funding_null_reason": null,
   "products_partnerships": "The tool is built on AF2-Multimer and publishes its CASP16-CAPRI dataset; it trades compute for accuracy rather than changing the model.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-af2-multimer",
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "It shows sampling volume raises the ceiling on the hardest interfaces; the bottleneck is selection.",
   "A sampling tool is not a new model; do not rank it alongside co-folding models.",
   "Thousands of structures per target cost far more compute than a single prediction.",
   "No official placing is reported; participation itself is the strongest evidence available."
  ],
  "caveats": [
   "The CASP line rests on the site's own page and is shown as unverified.",
   "The first release had been recorded as the data paper's date and is corrected to the tool preprint.",
   "Inference uses the official AF2 weights, so accuracy is capped by the underlying model."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s43588-024-00714-4",
    "title": "MassiveFold tool paper",
    "kind": "journal",
    "date": {
     "date": "2024-11-11",
     "precision": "day",
     "event_kind": "journal-publication",
     "basis": "Journal publication",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1038/s43588-024-00714-4",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://doi.org/10.1101/2025.05.26.653955",
    "title": "CASP16-CAPRI data paper preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1002/prot.70040",
    "title": "CASP16-CAPRI data paper, journal version",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/GBLille/MassiveFold/releases",
    "title": "MassiveFold GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://b2find.eudat.eu/dataset/00cc1668-0a62-5aed-8696-17cf75645a09",
    "title": "CASP16-CAPRI dataset",
    "kind": "archive",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-05 (the data paper preprint)",
    "now": "2024-04-30, the tool paper preprint; 2025-05 is the CASP16-CAPRI data paper",
    "source_url": "https://doi.org/10.1038/s43588-024-00714-4",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[1].source_url",
   "release_milestones.code_release.date"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-massivefold/",
   "en": "/bio/benchmark/models/massivefold/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-medgemma-med-gemini",
  "kind": "model",
  "short_slug": "medgemma-med-gemini",
  "names": {
   "display": "MedGemma / Med-Gemini",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "MedGemma",
    "Med-Gemini",
    "MedGemma 1.5"
   ],
   "ambiguous_aliases": [
    "Gemma"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "Med-Gemini",
     "id_or_null": null,
     "first_release": {
      "date": "2024-04-29",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "arXiv 2404.18416 v1 (no public weights)",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2404.18416",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "MedGemma 4B / 27B",
     "id_or_null": null,
     "first_release": {
      "date": "2025-07",
      "precision": "month",
      "event_kind": "preprint-posted",
      "basis": "Technical report arXiv 2507.05201; HF repo creation date not used",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2507.05201",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "MedGemma 1.5 4B",
     "id_or_null": null,
     "first_release": {
      "date": "2026-01-13",
      "precision": "day",
      "event_kind": "new-weights",
      "basis": "Official blog and weights release",
      "timestamp_utc": null,
      "source_url": "https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "Covers closed Med-Gemini (Gemini-based) and open-weight MedGemma versions (Gemma 3-based); dates and openness differ"
  },
  "family_null_reason": null,
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "Google medical language-model family: closed Med-Gemini and open-weight MedGemma",
  "org": {
   "names": [
    "Google Research",
    "Google DeepMind"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Medical multimodal LLM",
  "open_status": {
   "code": "partial",
   "weights": "open",
   "weights_license": "Health AI Developer Foundations terms (not Apache-2.0)",
   "code_license": null,
   "license_changed": null,
   "note": "MedGemma weights open under Google terms, not Apache-2.0; Med-Gemini closed"
  },
  "first_release": {
   "date": "2024-04-29",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv v1 of the earliest member, Med-Gemini",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2404.18416",
   "verification": "partial",
   "note": "This date is Med-Gemini (no public weights) while the latest update is MedGemma; read per member"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-04-29",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Med-Gemini arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2404.18416",
    "verification": "partial",
    "note": null
   },
   "journal_online": {
    "date": "2026-10-06",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "MedGemma published in Nature Medicine",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41591-026-04626-w",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": {
    "date": "2026-01-13",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "MedGemma 1.5 4B weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/google/medgemma-1.5-4b-it",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-10-06",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "MedGemma published in Nature Medicine",
   "description": "MedGemma paper in Nature Medicine; MedGemma 1.5 4B came out 2026-01-13; no MedGemma 2 found",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1038/s41591-026-04626-w",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "New weights and journal publication within 12 months",
  "what_it_does": "Google medical language models: MedGemma (4B multimodal, 27B text) is built on Gemma 3 with open weights; the earlier Med-Gemini is built on Gemini and closed. This site uses them mainly as biomedical QA comparators.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public medical text and images, including MedQA, MedMCQA and PubMedQA training splits",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "Larger internal non-public data, e.g. ~32.5M pathology patch-text pairs; licensing not stated",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-14-pubmedqa",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": null,
    "source_url": "https://arxiv.org/html/2507.05201",
    "confidence": "high",
    "verification": "verified",
    "note": "Developer-reported on the official protocol (MedGemma versions); Med-Gemini not reported"
   },
   {
    "benchmark_id": "bench-23-bioasq",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": null,
    "source_url": "https://ceur-ws.org/Vol-4283/paper1.pdf",
    "confidence": "high",
    "verification": "unverified",
    "note": "The BioASQ overview ranks systems, with no single-model MedGemma result; the site matrix entry is unsupported"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "Google Research and Google DeepMind (merged 2023-04) are part of Alphabet; MedGemma belongs to the Health AI Developer Foundations program",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": null,
   "products_partnerships_null_reason": "No product or partnership found",
   "related_models": [
    "model-txgemma",
    "model-c2s-scale",
    "model-frontier-llms"
   ]
  },
  "reading_notes": [
   "First release and latest update belong to different members; not one model's timeline",
   "The BioASQ pairing is unconfirmed; only PubMedQA counts"
  ],
  "caveats": [
   "Family entry: Med-Gemini is closed with no weights; MedGemma has open weights",
   "MedGemma training includes PubMedQA training splits; read its numbers with that in mind"
  ],
  "sources": [
   {
    "url": "https://api.crossref.org/works/10.1038/s41591-026-04626-w",
    "title": "Crossref record 10.1038/s41591-026-04626-w",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2507.05201",
    "title": "arXiv preprint 2507.05201",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/google/medgemma-1.5-4b-it",
    "title": "Hugging Face page google/medgemma-1.5-4b-it",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/",
    "title": "Official page research.google",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2404.18416",
    "title": "arXiv preprint 2404.18416",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release",
   "benchmarks_reported[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-medgemma-med-gemini/",
   "en": "/bio/benchmark/models/medgemma-med-gemini/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-minimol",
  "kind": "model",
  "short_slug": "minimol",
  "names": {
   "display": "MiniMol",
   "full": null,
   "aliases": [
    "MiniMol",
    "minimol"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single model; the group's earlier MolE has its own record",
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "Ten-million-parameter molecular foundation model densely pretrained on thousands of supervised tasks",
  "org": {
   "names": [
    "Valence Labs (Recursion Pharmaceuticals)"
   ],
   "global_entity_ids": [
    "recursion-pharmaceuticals"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (molecular foundation model, graph model)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The authors call it open source and a public package is installable; this round the official repository and weights were not found under the expected organisation path (the old path is dead), so licence and weight status are unverified"
  },
  "first_release": {
   "date": "2024-04-23",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2404.14986 v1 submission date (also accepted at a machine-learning conference workshop)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2404.14986",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-04-23",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2404.14986",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-04-23",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 (no later version)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2404.14986",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-08-22",
    "precision": "day",
    "year_hint": null,
    "basis": "Package registry 1.3.5 (1.3.4 on 2024-08-05)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://pypi.org/project/minimol/",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2024-08-22",
   "precision": "day",
   "year_hint": null,
   "basis": "Public package release 1.3.5 on the registry",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://pypi.org/project/minimol/",
   "verification": "verified",
   "description": "Public package releases 1.3.4 (2024-08-05) and 1.3.5 (2024-08-22); the matching repository exists with no releases or tags and a last commit of 2025-05-29 for storage-path maintenance. This is the latest sourced event",
   "note": "Corrects the site's \"no update since release\" wording: these package versions post-date the first release"
  },
  "status": "maintained",
  "status_note": "Maintenance commits only, the last on 2025-05-29",
  "what_it_does": "Takes the small-model, many-supervised-tasks route: about ten million parameters pretrained on millions of molecules, hundreds of millions of labels and roughly three thousand sparsely defined graph- and node-level tasks spanning quantum chemistry and biology, then applied to downstream property prediction. Against the group's earlier MolE it uses a smaller molecule set with denser labels, and against Uni-Mol2 it is more than two orders of magnitude smaller, emphasising parameter efficiency.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 6 million molecules, 500 million labels and 3,300 graph- and node-level tasks; the abstract lists no database names",
    "source_url": "https://arxiv.org/abs/2404.14986"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://arxiv.org/abs/2404.14986",
    "confidence": "high",
    "second_hand": false,
    "note": "ADMET group (22 datasets): the abstract claims significant gains over the previous best foundation model on 17 of them without naming the comparison model; official per-task board entries also exist, placing behind several fingerprint and gradient-boosting methods"
   }
  ],
  "other_evaluations": [
   {
    "label": "Parameter-efficiency comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "Uses a far smaller parameter count than peer foundation models as its selling point and comparison; a self-reported framing, not an independent evaluation",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: dense multitask supervised pretraining lets a ten-million-parameter model significantly beat the previous best molecular foundation model on most ADMET tasks.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The public per-task ADMET board carries its entries so they can be read alongside other submissions, though the organisers did not re-run the model independently.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The developer is the AI research arm of a listed AI-pharma company, announced in 2023 at a machine-learning conference and formed by merging the parent's machine-learning team with an acquired molecular-design company; the parent was founded 2013-11-05 and is based in Salt Lake City, Utah.",
   "funding": null,
   "products_partnerships": "The research arm co-developed the co-folding and affinity model Boltz-2 with a US university team; its earlier in-house model is MolE.",
   "related_models": [
    "model-mole",
    "model-boltz-2",
    "model-uni-mol2",
    "model-kermt"
   ]
  },
  "reading_notes": [
   "\"Better on 17 tasks\" is a comparison with the previous best foundation model, not with all methods: on the official per-task board several fingerprint and gradient-boosting models rank higher.",
   "It is a workshop paper with no journal version and only a v1 preprint.",
   "The benchmark's datasets are small with high variance, so rank gaps are close to noise.",
   "The official repository and weights were not located this round, so openness is treated as unverified."
  ],
  "caveats": [
   "Open-source and weight status unverified (the expected repository path is dead).",
   "The abstract does not name the foundation model it is compared against.",
   "Workshop paper only, no journal version.",
   "The site's \"no update since release\" wording is corrected to the public package releases."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2404.14986",
    "title": "Preprint arXiv 2404.14986",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://icml.cc/virtual/2024/35909",
    "title": "Conference workshop entry",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://pypi.org/project/minimol/",
    "title": "Package registry page",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://github.com/graphcore-research/minimol",
    "title": "Code repository (no releases or tags)",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://tdcommons.ai/benchmark/admet_group/01caco2/",
    "title": "Public ADMET per-task board (sample task)",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2024-04-23, no update since release",
    "now": "2024-08-22, public package 1.3.5 (overturns \"no later event\")",
    "source_url": "https://pypi.org/project/minimol/",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "none-since-release",
    "now": "code-release",
    "source_url": "https://pypi.org/project/minimol/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.code",
   "open_status.weights",
   "open_status.code_license",
   "open_status.weights_license",
   "training_data.b_private_licensed"
  ],
  "as_of": "2026-11-16",
  "global_entity": "recursion-pharmaceuticals",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-minimol/",
   "en": "/bio/benchmark/models/minimol/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-mole",
  "kind": "model",
  "short_slug": "mole",
  "names": {
   "display": "MolE",
   "full": null,
   "aliases": [
    "MolE"
   ],
   "ambiguous_aliases": [
    "MolE model",
    "Mole"
   ]
  },
  "family": null,
  "family_null_reason": "The sibling successor MiniMol has its own record; no family entry",
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "Molecular-graph foundation model adapting disentangled attention to graphs for property prediction",
  "org": {
   "names": [
    "Recursion"
   ],
   "global_entity_ids": [
    "recursion-pharmaceuticals"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (molecular-graph foundation model)",
  "open_status": {
   "code": "open",
   "weights": "closed",
   "weights_license": null,
   "code_license": "CC-BY-NC",
   "license_changed": null,
   "note": "Code is open for non-commercial use and the full pretrained weights are not released; a third-party container offers a smaller pretrained checkpoint"
  },
  "first_release": {
   "date": "2022-11-03",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2211.02657 v1 submitted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2211.02657",
   "verification": "verified",
   "note": "The site treated the 2024-11 journal publication as the release; first release follows the preprint v1"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2022-11-03",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2211.02657",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2022-11-03",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2211.02657",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2024-11-12",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal publication (DOI 10.1038/s41467-024-53751-y)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41467-024-53751-y",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2024-10-04",
    "precision": "day",
    "year_hint": null,
    "basis": "First commit in the public code repository",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/recursionpharma/mole_public",
    "verification": "partial",
    "note": "Last commit around 2025-06 per repository badges"
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2024-11-12",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41467-024-53751-y",
   "verification": "verified",
   "description": "Journal publication; the public code repository's first commit was 2024-10-04 and the last around 2025-06. This is the latest sourced substantive event",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only maintenance commits since publication, and a successor model exists",
  "what_it_does": "Works directly on molecular graphs with atom identities as tokens and graph topological distance as relative position, pretraining first self-supervised (masking atoms and predicting their neighbourhood) and then with graph-level supervision, for downstream property and activity prediction. The sibling successor MiniMol self-reports better results on most comparable tasks with roughly a tenth of the parameters, which makes this model a reference point alongside ADMET-AI and KERMT in ADMET work.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Self-supervised on about 842 million molecular graphs from a public compound library (with smaller subsets as controls), then supervised pretraining on about 456,000 molecules with labels from a public bioactivity database",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11557931/"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11557931/"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The paper reports no wet-lab validation",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11557931/"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://doi.org/10.1038/s41467-024-53751-y",
    "confidence": "medium",
    "second_hand": false,
    "note": "The pairing file records an official board relation, but this round's full sweep of the 22 official sub-task leaderboards found no row for the model: the journal version compares against a 2023-09 board snapshot and must not be read as a current board position, so this row is shown as unverified"
   }
  ],
  "other_evaluations": [
   {
    "label": "Pretraining scale ablation",
    "category": "own-task-set",
    "detail": "Controls on smaller molecular-graph subsets assessing how self-supervised scale affects downstream tasks; author-reported",
    "source_type": "self"
   },
   {
    "label": "Comparison against classical baselines",
    "category": "comparison-vs-baseline-only",
    "detail": "On some tasks fingerprint plus gradient-boosting baselines remain competitive, a point made both by the authors and by later work",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: moving disentangled attention to molecular graphs with two-stage pretraining beat the then-published best results on most tasks of a public ADMET task group (the comparison is a 2023-09 board snapshot).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent third-party evaluation was found; the successor's comparison comes from the same organisation and is not independent",
  "background": {
   "org_profile": "The developer is an AI-native biotech founded in Salt Lake City in 2013-11 and listed on Nasdaq in 2021-04, grown out of machine learning on high-content cell microscopy; it completed the acquisition of a UK AI drug-design company on 2024-11-20, after which headcount figures differ by source (roughly 600 to 800). Its machine-learning research centre launched 2023-07.",
   "funding": "A listed company. In 2023-07 a GPU vendor invested $50 million through a private placement alongside a foundation-model collaboration.",
   "products_partnerships": "Drug-discovery collaborations with two large pharma companies (from 2021-12 and 2023-09; announcements through 2026-08 report over $200 million in upfront and milestone payments received). A trade outlet says this model was released jointly with a Danish pharma company while the paper's affiliations read differently; the two accounts do not agree. A new chief executive took office 2026-01-01.",
   "related_models": [
    "model-minimol",
    "model-admet-ai",
    "model-kermt"
   ]
  },
  "reading_notes": [
   "The \"first on most tasks\" claim belongs to its 2024 paper and a 2023-09 board snapshot only: this round's sweep of all 22 official sub-task leaderboards found no row for the model, so it cannot be presented as a current board position.",
   "Preprint and journal versions count differently, showing how sensitive such rankings are to the snapshot date.",
   "The full weights are not released, so outsiders cannot reproduce the whole model and must use a third-party smaller checkpoint.",
   "On some TDC tasks classical fingerprint methods remain competitive, so a task-group ranking is not a general advantage."
  ],
  "caveats": [
   "No row for the model was found on the official sub-task leaderboards, so the board relation is shown as unverified.",
   "The comparison is a 2023-09 board snapshot and has since been surpassed.",
   "The full pretrained weights are not released.",
   "Media and paper accounts disagree on which organisations co-released the model."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41467-024-53751-y",
    "title": "Journal paper 2024",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11557931/",
    "title": "Journal full text (public archive)",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2211.02657",
    "title": "Preprint v1",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/recursionpharma/mole_public",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "title": "Official task-group leaderboard overview (basis of this round's check)",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "First on 10 of 22 tasks (stated as a leaderboard position)",
    "now": "The claim belongs only to its 2024 paper and a 2023-09 snapshot; a full scan of the 22 official per-task boards found no row for it, so it is not a current leaderboard position",
    "source_url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2024-11 (journal publication)",
    "now": "2022-11-03（arXiv v1）",
    "source_url": "https://arxiv.org/abs/2211.02657",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[0]",
   "release_milestones.code_release",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-mole/",
   "en": "/bio/benchmark/models/mole/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developer company but none for this model",
  "status_head": null
 },
 {
  "id": "model-molformer-xl",
  "kind": "model",
  "short_slug": "molformer-xl",
  "names": {
   "display": "MoLFormer-XL",
   "full": null,
   "aliases": [
    "MoLFormer-XL",
    "MoLFormer-XL-both-10pct"
   ],
   "ambiguous_aliases": [
    "MoLFormer"
   ]
  },
  "family": null,
  "family_null_reason": "The later generative variant from the same organisation has no site entity record, so no family entry",
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "Chemical language model pretraining on billion-scale molecule strings, then fine-tuned for property prediction",
  "org": {
   "names": [
    "IBM Research"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Property prediction (molecule-string language model)",
  "open_status": {
   "code": "open",
   "weights": "partial",
   "weights_license": "Apache-2.0 (model card)",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code is open; the public weights are a subset model trained on 10% of each corpus, and the full large-model weights are not released"
  },
  "first_release": {
   "date": "2021-06-17",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2106.09553 v1 submitted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2106.09553",
   "verification": "verified",
   "note": "The site used the 2022 journal year as the release; first release follows the preprint v1"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2021-06-17",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2106.09553",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2021-06-17",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1 (figures follow the v3 full text)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2106.09553",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2022-12-21",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal publication (DOI 10.1038/s42256-022-00580-7)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s42256-022-00580-7",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": null,
    "precision": null,
    "year_hint": 2022,
    "basis": "Public code repository; the first public date was not retrieved this round and the last commit is 2023-10-16 (checkpoint usage documentation)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/IBM/molformer",
    "verification": "unverified",
    "note": null
   },
   "weights_release": {
    "date": null,
    "precision": null,
    "year_hint": null,
    "basis": "Only a 10% subset checkpoint is public; its upload date could not be retrieved from the host's interface",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/ibm/MoLFormer-XL-both-10pct",
    "verification": "unverified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2022-12-21",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s42256-022-00580-7",
   "verification": "verified",
   "description": "Journal publication; the code repository's last commit is 2023-10-16 (checkpoint usage documentation). This is the latest sourced substantive event",
   "note": null
  },
  "status": "maintained",
  "status_note": "No substantive event after 2023-10, though it is still widely used as a baseline",
  "what_it_does": "Treats molecule strings as text, using a twelve-layer Transformer with linear attention and rotary position embeddings so that billion-scale pretraining is feasible, then fine-tunes on downstream property tasks; the paper argues the representation captures molecular geometry and structure. Unlike graph models such as MolE it needs only strings rather than graphs or conformations, and unlike Uni-Mol2 it does not depend on 3D conformations.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Up to 1.1 billion molecule strings from two public compound libraries; the public version uses 10% of each",
    "source_url": "https://doi.org/10.1038/s42256-022-00580-7"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1038/s42256-022-00580-7"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The paper reports no wet-lab validation",
    "source_url": "https://doi.org/10.1038/s42256-022-00580-7"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-03-moleculenet",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2106.09553",
    "confidence": "high",
    "second_hand": false,
    "note": "Six classification tasks on scaffold splits and five regression tasks on random splits, eleven in all; the comparison baselines are supervised graph neural networks with no fingerprint baseline. No figures are transcribed here"
   }
  ],
  "other_evaluations": [
   {
    "label": "Geometry probe of the representation",
    "category": "own-task-set",
    "detail": "The authors use self-built analyses to argue the string representation carries geometric and structural information",
    "source_type": "self"
   },
   {
    "label": "Corpus-scale controls",
    "category": "own-task-set",
    "detail": "Controls across pretraining corpus fractions; the public release corresponds to the 10% subset setting",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: linear attention makes billion-scale string pretraining feasible, and after fine-tuning it beats the supervised graph-neural-network baselines of the time on public property tasks.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evaluation of the model was found, and task-group numbers in third-party papers are not board results",
  "background": {
   "org_profile": "The developer is the research division of a large US technology company, centred in New York State; its careers page claims over three thousand researchers across twelve labs on six continents (the page is undated). A change of research director appears only in second-hand sources with inconsistent months and is not concluded here.",
   "funding": null,
   "products_partnerships": "The company's chemistry foundation-model line includes this model, a later decoder-only generative model and a target-specific generative model, plus a public molecular foundation-model site; on-site comparisons are MolE (graph) and Uni-Mol2 (3D).",
   "related_models": [
    "model-mole",
    "model-uni-mol2",
    "model-minimol"
   ]
  },
  "reading_notes": [
   "These are 2022 results that later foundation models use as a baseline, yet the full large-model weights are not public, so independent reproduction is hard.",
   "MoleculeNet datasets are small and split-sensitive, and studies report that most pretrained embeddings do not beat classical fingerprints.",
   "None of the 22 dataset pages of the official TDC ADMET board carries an entry for this model (checked 2026-11-16), so task-group numbers in third-party papers are not board results.",
   "The comparison baselines are weak (supervised graph networks only), so cross-paper comparisons need fingerprint baselines added."
  ],
  "caveats": [
   "The full weights are not public; only the 10% subset model is available.",
   "First dates for code and weights could not be retrieved this round and are marked unverified.",
   "All benchmark results are author-reported with no board entry.",
   "No substantive update after 2023-10."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s42256-022-00580-7",
    "title": "Journal paper 2022",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2106.09553",
    "title": "Preprint (figures follow v3)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/IBM/molformer",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/ibm/MoLFormer-XL-both-10pct",
    "title": "Public subset weights and model card",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "title": "Another task-group leaderboard overview (used to rule out a pairing)",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2022 (journal publication year)",
    "now": "2021-06-17（arXiv v1）",
    "source_url": "https://arxiv.org/abs/2106.09553",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Candidate pairing TDC ADMET task group",
    "now": "Not a pairing: none of the 22 dataset pages on the official board has an entry for this model",
    "source_url": "https://tdcommons.ai/benchmark/admet_group/overview/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.code_release",
   "release_milestones.weights_release",
   "background.org_profile"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-molformer-xl/",
   "en": "/bio/benchmark/models/molformer-xl/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-mosaic",
  "kind": "model",
  "short_slug": "mosaic",
  "names": {
   "display": "Mosaic",
   "full": null,
   "aliases": [],
   "ambiguous_aliases": [
    "Mosaic"
   ]
  },
  "family": null,
  "family_null_reason": "A single open-source library with no family members",
  "domain": "protein",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Open multi-objective protein design library optimising binders through combined scorers",
  "org": {
   "names": [
    "Escalante Bio"
   ],
   "global_entity_ids": []
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (multi-objective differentiable optimisation library)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": "MIT",
   "license_changed": null,
   "note": "The library has no weights of its own and depends on third-party scorer weights under their own terms"
  },
  "first_release": {
   "date": "2025-01-12",
   "precision": "day",
   "year_hint": null,
   "basis": "Repository creation and first commit (no earlier public announcement or paper found this round)",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/escalante-bio/mosaic",
   "verification": "unverified",
   "note": "Policy forbids repository creation as a first release, so this date must display as unverified; the only tagged version is v0.1.0 (2026-10-06)"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-01-12",
    "precision": "day",
    "year_hint": null,
    "basis": "First repository commit (not admissible as first-release evidence)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/escalante-bio/mosaic",
    "verification": "unverified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2026-10-06",
    "precision": "day",
    "year_hint": null,
    "basis": "First tagged version v0.1.0 (for an archive deposit)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://api.github.com/repos/escalante-bio/mosaic/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-11-05",
   "precision": "day",
   "year_hint": null,
   "basis": "Repository commits: structure-model contract and sampler options",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://api.github.com/repos/escalante-bio/mosaic/commits",
   "verification": "verified",
   "description": "Adds a structure-model contract and exponential/geometric sampler options; recent commits added two structure backends and an enzyme-design backend and removed an older one; v0.1.0 tagged 2026-10-06",
   "note": null
  },
  "status": "active",
  "status_note": "Tagged release and ongoing commits within 12 months",
  "what_it_does": "Not a single model but a library that combines co-folding predictors and other scoring terms into one differentiable multi-objective loss, then optimises sequences by gradient to design binders, with swappable scorers that track newer models. Like BindCraft it backpropagates through scorers, but its backends are interchangeable; it differs from the trained generative route of BoltzGen, and typical backends come from models such as Boltz-2.",
  "training_data": {
   "a_public": {
    "value": "no",
    "detail": "No training of its own; relies on the weights of the models it calls",
    "source_url": "https://github.com/escalante-bio/mosaic"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Designs assayed by a protein measurement service provider, for validation only",
    "source_url": "https://blog.escalante.bio/minibinder-design-is-just-not-that-hard/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Competition entry (de novo track)",
    "targets": 1,
    "designs_tested": 10,
    "hits": 9,
    "denominator_basis": "Ten designs, one target, self-reported",
    "success_rate": "9/10",
    "affinity_best": null,
    "version": "developer blog",
    "source_type": "self",
    "source_url": "https://blog.escalante.bio/winning-the-de-novo-portion-of-the-adaptyv-nipah-binder-competition/",
    "date": {
     "date": "2026-01-29",
     "precision": "day",
     "year_hint": null,
     "basis": "Developer blog post",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://blog.escalante.bio/winning-the-de-novo-portion-of-the-adaptyv-nipah-binder-competition/",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Executed by the organiser and its measurement provider; hit counts self-reported",
    "unverified_marker": null,
    "note": "The \"1st-4th\" framing means four of its designs ranked top-four de novo by affinity, not participant placings; the organiser's participant ranking differs. Keep the n=10 single-target caveat"
   },
   {
    "campaign": "Own minibinder tests (two targets)",
    "targets": 2,
    "designs_tested": 60,
    "hits": null,
    "denominator_basis": "Ten designs per method per target, three methods across two targets, 60 sent in total",
    "success_rate": "8/10 and 7/10 (best of three design arms)",
    "affinity_best": null,
    "version": "developer blog",
    "source_type": "self",
    "source_url": "https://blog.escalante.bio/minibinder-design-is-just-not-that-hard/",
    "date": {
     "date": "2026",
     "precision": "year",
     "year_hint": null,
     "basis": "Developer blog; the page gives no sourced date",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://blog.escalante.bio/minibinder-design-is-just-not-that-hard/",
     "verification": "unverified",
     "note": null
    },
    "verification": "verified",
    "executor": "Executed by a protein measurement service provider",
    "unverified_marker": null,
    "note": "Both rates are the best method arm per target and must be stated as selected across three arms"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-28-adaptyv",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://proteinbase.com/collections/nipah-binder-competition-results",
    "confidence": "high",
    "second_hand": false,
    "note": "Entered the competition's de novo track; the results page lists designs rather than per-tool hit rates, and hit counts are self-reported"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own two-target minibinder test",
    "category": "wet-lab-validation",
    "detail": "Three design arms across two targets, reporting the best arm's per-target hit rate",
    "source_type": "self"
   },
   {
    "label": "Scorer interchangeability demonstration",
    "category": "case-study",
    "detail": "The repository demonstrates plugging different structure and enzyme-design backends into the same optimisation loop",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: assembling off-the-shelf scorers into a differentiable objective already yields minibinders that bind at a high rate, with top-ranked designs in a public competition's de novo track.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The organiser's results page records design-level outcomes for that competition and can be checked against the entry, but it gives no per-tool hit rate and so is not an independent evaluation of the library.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "A small protein-design company focused on computational minibinder design and open-source tooling; its jobs page calls it a well-funded startup, while founding year, funding and headcount are not publicly disclosed and so are not stated.",
   "funding": null,
   "products_partnerships": "The product is the open library itself, alongside open work porting an open binder-generation model to another framework (per a developer social post, second-hand); no pharma partnerships are disclosed.",
   "related_models": [
    "model-bindcraft",
    "model-boltz-2",
    "model-boltzgen"
   ]
  },
  "reading_notes": [
   "The 9/10 figure rests on ten designs against one target, so the confidence interval is wide.",
   "Distinguish the ranking: four of its designs placed top-four, which is not a participant placing.",
   "The 8/10 and 7/10 figures are best-of-three arms and must not be read as one method's hit rate.",
   "It is a library, not a model, so performance tracks the chosen scorers; its first-release date is inadmissible under policy and displays as unverified."
  ],
  "caveats": [
   "The first-release basis is repository creation or first commit, which policy disallows; it must show as unverified.",
   "Hit rates and placings come from the developer blog, and the competition ranks by affinity.",
   "The library has no training or weights of its own, so performance cannot be attributed to it alone.",
   "The self-test blog page carries no sourced date."
  ],
  "sources": [
   {
    "url": "https://github.com/escalante-bio/mosaic",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://api.github.com/repos/escalante-bio/mosaic/releases",
    "title": "Releases API",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://blog.escalante.bio/winning-the-de-novo-portion-of-the-adaptyv-nipah-binder-competition/",
    "title": "Developer competition blog post",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://blog.escalante.bio/minibinder-design-is-just-not-that-hard/",
    "title": "Developer minibinder blog post",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://proteinbase.com/collections/nipah-binder-competition-results",
    "title": "Competition results page",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "Took first to fourth place in de novo (participant placings)",
    "now": "Its four designs rank top four by affinity; the organizer's participant ranking is different",
    "source_url": "https://blog.escalante.bio/winning-the-de-novo-portion-of-the-adaptyv-nipah-binder-competition/",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Hit rates of 8/10 and 7/10 on two targets",
    "now": "Same numbers, but it must be noted that they come from the best of three design arms",
    "source_url": "https://blog.escalante.bio/minibinder-design-is-just-not-that-hard/",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2025 (no basis)",
    "now": "2025-01-12, based on repository creation / first commit, which the policy does not accept; shown as unverified",
    "source_url": "https://github.com/escalante-bio/mosaic",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.first_public_evidence",
   "wetlab_validation[1].date",
   "background.products_partnerships",
   "open_status.weights"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-mosaic/",
   "en": "/bio/benchmark/models/mosaic/"
  },
  "names_full_null_reason": "The repository uses the tool name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-nesso-1",
  "kind": "model",
  "short_slug": "nesso-1",
  "names": {
   "display": "Nesso-1",
   "full": null,
   "aliases": [
    "Nesso-1"
   ],
   "ambiguous_aliases": [
    "Nesso"
   ]
  },
  "family": null,
  "family_null_reason": "A single version with no family members",
  "domain": "small-molecule",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Open protein-ligand affinity model trading coarse-grained structure for inference speed",
  "org": {
   "names": [
    "Valence Labs",
    "Recursion Pharmaceuticals"
   ],
   "global_entity_ids": [
    "recursion-pharmaceuticals"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US",
   "CA"
  ],
  "model_class": "Structure-to-affinity (coarse-grained co-folding affinity)",
  "open_status": {
   "code": "unclear",
   "weights": "open",
   "weights_license": "Apache-2.0 per model card",
   "code_license": null,
   "license_changed": null,
   "note": "Weights are public on a model host with an Apache-2.0 model card; no separate code repository was found"
  },
  "first_release": {
   "date": "2026-07-11",
   "precision": "day",
   "year_hint": null,
   "basis": "The \"Publish v1.0.0\" commit on the weights host; a technical report PDF appeared the same month with no exact date found",
   "event_kind": "new-weights",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/api/models/recursionpharma/nesso/commits/main",
   "verification": "partial",
   "note": "Repository creation is inadmissible as a first release; the weights-publication commit of the same day is used instead, and the report's exact date is still missing"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-07-11",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights publication commit",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/api/models/recursionpharma/nesso/commits/main",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2026-08-03",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint posted date (registry record; the 08-01 in the DOI string is not used as a date)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.64898/2026.08.01.742196",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2026-07-11",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights host publication commit",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/recursionpharma/nesso",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-08-03",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint posted (registry posted date)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.64898/2026.08.01.742196",
   "verification": "verified",
   "description": "The preprint went up on 2026-08-03; later activity on the weights host is model-card and asset edits only (config 2026-07-28, figures 2026-08-06)",
   "note": "The kind previously read \"no update since release\", which contradicted the description; corrected to preprint posted"
  },
  "status": "active",
  "status_note": "Weights release and preprint within 12 months",
  "what_it_does": "Reuses the public training data and optimisation framework of Boltz-2 but drops its atom-level diffusion module, trading coarse-grained poses for inference speed in the same spirit as TerraBind; it outputs protein-ligand binding affinity and is positioned as a prefilter for large virtual screens.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Reuses the public training data and optimisation procedure of Boltz-2",
    "source_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No own wet-lab training data",
    "source_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Internal biochemical assays are held-out evaluation on existing data rather than new wet-lab work; the scale in internal projects is not disclosed",
    "source_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf",
    "confidence": "high",
    "second_hand": false,
    "note": "The report names the benchmark while copying baselines from another paper and conceding different filtering, so it is self-reported on a modified protocol"
   },
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf",
    "confidence": "high",
    "second_hand": false,
    "note": "Named in the same sentence; poses are coarse-grained so the physical-validity checks are not run as on all-atom predictions"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf",
    "confidence": "high",
    "second_hand": false,
    "note": "The third structural benchmark named in the same sentence, with the same metrics, baseline sourcing and filtering caveat"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf",
    "confidence": "high",
    "second_hand": false,
    "note": "Post-hoc use of a released round's affinity set (140 complexes, two targets), not participation in the blind round"
   }
  ],
  "other_evaluations": [
   {
    "label": "Public affinity benchmarks (off-site)",
    "category": "public-benchmark-not-on-site",
    "detail": "Correlation reported on a viral-protease compound set and several free-energy comparison sets, with baselines largely copied from other papers",
    "source_type": "self"
   },
   {
    "label": "Virtual-screening activity benchmark (off-site)",
    "category": "public-benchmark-not-on-site",
    "detail": "Average precision and enrichment reported, with baselines taken from the comparison model's paper",
    "source_type": "self"
   },
   {
    "label": "25 internal biochemical assays",
    "category": "own-task-set",
    "detail": "Compound-weighted correlation comparison; the report concedes limited gains on some assays",
    "source_type": "self"
   },
   {
    "label": "Inference speed comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "Throughput compared with the baseline on three internal targets of ten compounds each on a single accelerator; a higher multiple circulates in company remarks (second-hand)",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: an open coarse-grained co-folding affinity model more than an order of magnitude faster than the baseline, matching or beating it on their chosen public affinity and structure benchmarks, while cautioning that these are largely in-distribution with limited zero-shot generalisation.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation found; all figures come from the technical report and most baselines are copied",
  "background": {
   "org_profile": "The developer is the AI research unit of a listed US AI drug-discovery company that merged with a UK AI drug-design company in 2024 and has had a new chief executive since 2026-01-01 (the previous one moved to chair, then to strategic adviser from 2026-04).",
   "funding": null,
   "products_partnerships": "The company co-released the predecessor co-folding affinity model with a US university (2025-06), whose team later spun out as a public benefit corporation (see the BoltzGen record); this model is a slimmed, faster version of that predecessor and explicitly draws on TerraBind.",
   "related_models": [
    "model-boltz-2",
    "model-terrabind",
    "model-boltzgen",
    "model-aqaffinity"
   ]
  },
  "reading_notes": [
   "The headline is speed rather than an accuracy breakthrough, which suits prefiltering in virtual screens.",
   "All three structural benchmark entries are self-reported on a modified protocol: filtering admittedly differs from the cited paper and the poses are coarse-grained.",
   "The CASP entry is post-hoc reuse of a released affinity set, not participation, and does not count toward reported benchmarks.",
   "The report concedes limited generalisation and notes that on one benchmark a naive molecular-weight baseline is almost as good, so results should not be extrapolated to new targets."
  ],
  "caveats": [
   "All figures are self-reported with baselines mostly copied from other papers and no third-party recomputation.",
   "The first-release basis partly rests on repository creation and the report's exact date is missing; partly verified.",
   "The speed multiple has two framings (technical report and company remarks), the latter second-hand only.",
   "The scale of internal assay use is not disclosed."
  ],
  "sources": [
   {
    "url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf",
    "title": "Technical report PDF",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/recursionpharma/nesso",
    "title": "Weights and model card",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.64898/2026.08.01.742196",
    "title": "Preprint registry record",
    "kind": "archive",
    "date": null
   },
   {
    "url": "https://huggingface.co/api/models/recursionpharma/nesso/commits/main",
    "title": "Weights host commit history",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update_kind",
    "was": "No later sourced event found",
    "now": "Preprint online (2026-08-03)",
    "source_url": "https://api.crossref.org/works/10.64898/2026.08.01.742196",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Zero pairings (no pairing among the original 30)",
    "now": "Three structure benchmarks are modified-protocol self-reports and one is retrospective dataset use; the model leaves the zero-pairing list",
    "source_url": "https://www.valencelabs.com/wp-content/uploads/2026/07/nesso1.pdf",
    "date_found": "2026-11-16"
   },
   {
    "field": "release_milestones.paper_v1",
    "was": "Preprint 2026-08-01 (DOI string)",
    "now": "2026-08-03 (posted date from the registration agency)",
    "source_url": "https://api.crossref.org/works/10.64898/2026.08.01.742196",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.first_public_evidence",
   "release_milestones.weights_release",
   "training_data.b_private_licensed",
   "open_status.code",
   "other_evaluations[3]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-nesso-1/",
   "en": "/bio/benchmark/models/nesso-1/"
  },
  "names_full_null_reason": "The technical report gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 3,
   "by_relation": {
    "self-reported-modified-protocol": 3,
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the company but not for the model",
  "status_head": null
 },
 {
  "id": "model-neuralplexer3",
  "kind": "model",
  "short_slug": "neuralplexer3",
  "names": {
   "display": "NeuralPLexer3",
   "full": null,
   "aliases": [
    "NeuralPLexer3",
    "NP3"
   ],
   "ambiguous_aliases": [
    "NeuralPLexer"
   ]
  },
  "family": null,
  "family_null_reason": "Earlier NeuralPLexer 1 and 2 fall outside this record, which covers the third generation only",
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "Flow-matching co-folding model emphasising stereochemistry and physical validity",
  "org": {
   "names": [
    "Iambic Therapeutics"
   ],
   "global_entity_ids": [
    "iambic-therapeutics"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding (flow matching)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": "BSD-3-Clause",
   "license_changed": null,
   "note": "Weights and main code are not released; only the NPBench evaluation set is open under BSD-3"
  },
  "first_release": {
   "date": "2024-11-01",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer's NP3 Beta preview post",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.iambic.ai/post/np3-preview",
   "verification": "partial",
   "note": "The site recorded the arXiv v1 date, 2024-12-14, as first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-11-01",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer Beta preview post",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.iambic.ai/post/np3-preview",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2024-12-14",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2412.10743",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-12",
    "precision": "month",
    "year_hint": null,
    "basis": "Conference proceedings entry (December 2025 meeting)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://proceedings.neurips.cc/paper_files/paper/2025/hash/0010031a1b4910aa67edbda26a705518-Abstract-Conference.html",
    "verification": "partial",
    "note": "The exact day is not pinned; month precision only"
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-11-01",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer preview post",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.iambic.ai/post/np3-preview",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2025-12",
   "precision": "month",
   "year_hint": null,
   "basis": "Main-track entry in the conference proceedings",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://proceedings.neurips.cc/paper_files/paper/2025/hash/0010031a1b4910aa67edbda26a705518-Abstract-Conference.html",
   "verification": "partial",
   "description": "Entered a machine-learning conference's main-track proceedings in December 2025; no weights or main code release was found.",
   "note": "The meeting date is month-level; the event is recorded as a proceedings publication"
  },
  "status": "active",
  "status_note": "A proceedings event within 12 months, but no code or weights release",
  "what_it_does": "Generates protein-ligand, protein-protein and peptide complex structures with flow matching, with the paper stressing stereochemistry and physical validity. It is one of two core models on the developer's platform, the other being the property-prediction model Enchant; it differs from AlphaFold3 in its flow architecture and its emphasis on physical validity, and outside users reach it only through partnerships.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "A protein structure archive (before 2020-09-01), about 270,000 distilled entries from a public protein set, and about 12.7 million quantum-chemistry conformers",
    "source_url": "https://arxiv.org/abs/2412.10743"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The paper does not mention the developer's private data",
    "source_url": "https://arxiv.org/abs/2412.10743"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2412.10743",
    "confidence": "high",
    "second_hand": false,
    "note": "Reported on the V2 set only, with RMSD under 2 Å plus the physical-validity check as the metric"
   },
   {
    "benchmark_id": "bench-25-rnapuzzles",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2412.10743v2",
    "confidence": "high",
    "second_hand": false,
    "note": "Uses an eight-target RNA subset from one CASP round, with baseline numbers copied from another paper rather than rerun"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://arxiv.org/html/2412.10743v2",
    "confidence": "medium",
    "second_hand": false,
    "note": "The CASP cell and the RNA cell share one piece of evidence; this is after-the-fact use of released data and remains to be rechecked"
   }
  ],
  "other_evaluations": [
   {
    "label": "Developer's own evaluation set",
    "category": "own-task-set",
    "detail": "A self-built set of 1,143 targets covering protein-protein interfaces, peptides and conformational change",
    "source_type": "self"
   },
   {
    "label": "Comparison with docking software",
    "category": "comparison-vs-baseline-only",
    "detail": "The paper places its results alongside two classical docking programs and one diffusion model",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a higher share of poses meeting both RMSD under 2 Å and physical validity than AlphaFold3 on the V2 pose set, near-perfect stereochemistry and second-scale inference.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Weights and main code are not public and no third-party rerun of the model was found",
  "background": {
   "org_profile": "The developer is an AI-native oncology drug company founded in 2019 in California by two chemistry professors from a US institute of technology and a UK university, growing out of a quantum-chemistry software company; its platform spans structure prediction, property prediction and synthesis planning. It listed on a US exchange in September 2026.",
   "funding": "A $100M Series B plus a $50M extension in June 2024; a September 2026 listing of 9.4 million shares at $16 raised about $150M, with the prospectus allocating roughly $20M to training new versions of its two core models.",
   "products_partnerships": "Its pipeline includes a Phase 1 oral HER2 inhibitor and two preclinical programmes; partners include a Japanese pharmaceutical company (reported milestones up to $1.7B), a US oncology company ($25M for custom training) and another large pharmaceutical company.",
   "related_models": [
    "model-alphafold3",
    "model-boltz-2",
    "model-enchant-v1-v2"
   ]
  },
  "reading_notes": [
   "The comparison with AlphaFold3 rests on numbers the paper recomputed itself; other papers report widely differing figures, so any citation needs its source.",
   "Without the physical-validity requirement the two are close, so the advantage sits in validity alone.",
   "With no public weights nothing can be rechecked externally; only the open evaluation set is reproducible.",
   "The RNA comparison uses eight targets with baselines quoted from another paper, so it carries little weight."
  ],
  "caveats": [
   "Weights and main code are unreleased; outside use runs through partnerships.",
   "First release is taken from the developer preview post, with the paper v1 later; both are recorded.",
   "The proceedings event is pinned only to the month of the meeting.",
   "The CASP pairing reuses the RNA evidence and remains to be rechecked.",
   "Own wet-lab training data is not disclosed, which is not the same as none."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2412.10743",
    "title": "Preprint 2024",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2412.10743v2",
    "title": "Preprint full text v2",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://proceedings.neurips.cc/paper_files/paper/2025/hash/0010031a1b4910aa67edbda26a705518-Abstract-Conference.html",
    "title": "Conference proceedings entry 2025",
    "kind": "paper",
    "date": null
   },
   {
    "url": "https://www.iambic.ai/post/np3-preview",
    "title": "Developer Beta preview post",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2024-12-14（arXiv v1）",
    "now": "2024-11-01 (developer beta preview post); arXiv v1 kept as paper_v1",
    "source_url": "https://www.iambic.ai/post/np3-preview",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.journal_online",
   "latest_update",
   "training_data.c_own_wetlab_for_training",
   "benchmarks_reported"
  ],
  "as_of": "2026-11-16",
  "global_entity": "iambic-therapeutics",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-neuralplexer3/",
   "en": "/bio/benchmark/models/neuralplexer3/"
  },
  "names_full_null_reason": "The name is a product name plus version; the paper gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 2,
   "by_relation": {
    "self-reported-official-protocol": 1,
    "self-reported-modified-protocol": 1,
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-ntv3",
  "kind": "model",
  "short_slug": "ntv3",
  "names": {
   "display": "NTv3",
   "full": "Nucleotide Transformer v3",
   "aliases": [
    "NTv3",
    "Nucleotide Transformer v3"
   ],
   "ambiguous_aliases": [
    "Nucleotide Transformer",
    "NT"
   ]
  },
  "family": null,
  "family_null_reason": "8M, 100M and 650M are sizes of one generation; the two earlier generations have no entity records",
  "domain": "genomics-rna",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Genome foundation model with 1 Mb context doing track prediction, gene annotation and sequence generation in one",
  "org": {
   "names": [
    "InstaDeep (subsidiary of BioNTech)"
   ],
   "global_entity_ids": []
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "GB"
  ],
  "model_class": "Genome language model / sequence-to-function model",
  "open_status": {
   "code": "open",
   "weights": "research-only",
   "weights_license": "custom licence (model hub label: other)",
   "code_license": null,
   "license_changed": null,
   "note": "Weights are on the model hub under a custom licence (checked 2026-11-16) with a JAX code release; the companion task suite is non-commercial"
  },
  "first_release": {
   "date": "2025-12-25",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1 posted date (version API); a technical report PDF appeared the same month",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
   "verification": "verified",
   "note": "The earlier 2025-12-22 was inferred from the DOI string, which this round forbids as a date"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-12-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-12-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-02-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog with code and weights the same day",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://instadeep.com/2026/02/modelling-the-genome-with-ntv3/",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-02-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Full model-family weights published to the model hub",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://instadeep.com/2026/02/modelling-the-genome-with-ntv3/",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-02-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://instadeep.com/2026/02/modelling-the-genome-with-ntv3/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-29",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v2 posted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
   "verification": "verified",
   "description": "Preprint v2 posted; the earlier event is the official blog and weights release (2026-02-18), which also published the developers' own multi-species task suite. Repository syncs through 2026-11-02 are maintenance",
   "note": "The latest update was recorded as the 2026-02-18 product update; corrected to the later paper revision"
  },
  "status": "active",
  "status_note": "Paper revision, weights release and repository maintenance within 12 months",
  "what_it_does": "Extends the two earlier generations from short-sequence classification to a 1 Mb context, with one model emitting functional tracks and gene-structure annotation and supporting conditional generation; post-training uses about 16,000 tracks and annotations across 24 species. It is far smaller than AlphaGenome but covers more species, and unlike Evo 2 it adds supervised track post-training.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Pretrained on a public pan-genome dataset (the report cites 9T bases trained) and post-trained on public functional-genomics tracks and annotations, including human and mouse track sets plus plant and insect data",
    "source_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Generated enhancers validated by reporter sequencing in one insect cell line, for validation only",
    "source_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Reporter-sequencing validation of generated enhancers",
    "targets": null,
    "designs_tested": 900,
    "hits": null,
    "denominator_basis": "900 sequences = 600 activity-bin designs (60 per bin x 5 bins x 2 promoters) + a 150 vs 150 head-to-head; one cell type",
    "success_rate": null,
    "affinity_best": null,
    "version": "technical report (not peer reviewed)",
    "source_type": "self",
    "source_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf",
    "date": {
     "date": "2025-12",
     "precision": "month",
     "year_hint": null,
     "basis": "Technical report (month-level evidence only)",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf",
     "verification": "partial",
     "note": null
    },
    "verification": "partial",
    "executor": "Developer-run and not peer reviewed",
    "unverified_marker": null,
    "note": "The source is internally inconsistent (one section says 1,000, the methods say 900); 900 is used per the methods and the discrepancy noted. No hit rate is reported, only binning and promoter-specificity ratios"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-08-enformer",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf",
    "confidence": "high",
    "second_hand": false,
    "note": "Fine-tuned on the predecessor track set's folds and settings; only relative gains are given and there is no direct comparison with the other big-lab model"
   }
  ],
  "other_evaluations": [
   {
    "label": "Developer-built multi-species task suite",
    "category": "own-task-set",
    "detail": "106 tasks across 7 species with base-resolution outputs; released alongside the model, with competitors chosen by the developers and the win list paraphrased from the blog",
    "source_type": "self"
   },
   {
    "label": "GUE / NT short-sequence suite",
    "category": "public-benchmark-not-on-site",
    "detail": "Not a pairing: the report does not use that suite and compares gene annotation on an earlier in-house benchmark instead; that benchmark entry's new version was released alongside this model's preprint",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one 1 Mb-context model covers track prediction, gene annotation and generation, beating several specialised tools and a larger sequence model on their own multi-species task suite.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation under a site benchmark protocol was found, and the in-house suite has no third-party re-run",
  "background": {
   "org_profile": "A decision-AI company founded in North Africa in 2014 and headquartered in London since 2015, with biology as a focus line; it became a subsidiary of an mRNA and immuno-oncology pharma company in 2023. At a 2025-10 event the developer said its nucleotide model family had over a million downloads and over five hundred citations. No citable official headcount, so none is given.",
   "funding": "A $7M Series A in 2019-05; a $100M Series B in 2022-01 (lead investor framings differ: one outlet says two funds co-led, another names an internet company); acquired by the parent in 2023-01 for about $389M upfront with up to about $243M in milestones, closing 2023-07-31.",
   "products_partnerships": "Builds on the family's two earlier nucleotide models (the previous generation was published in a journal in 2025-02); the developer also has a protein mass-spectrometry model, and the parent is an mRNA and immuno-oncology company. Three European and US research institutions collaborated. Comparators are Evo 2 and AlphaGenome.",
   "related_models": [
    "model-evo-2",
    "model-alphagenome",
    "model-borzoi"
   ]
  },
  "reading_notes": [
   "The 106-task suite was built by the developers and released with the model, so it only shows leadership on tasks they defined.",
   "There is no GUE/NT suite result; on the track set only relative gains are given and the other big-lab model is not compared directly.",
   "The largest size is only 650M, so an edge over a smaller competitor configuration cannot be extrapolated to that competitor's largest one.",
   "The wet lab covers about 900 sequences in one cell type and reports binning and specificity ratios rather than hit rates, with the source inconsistent on the count."
  ],
  "caveats": [
   "The main evaluation is a developer-built suite with no third-party re-run.",
   "The technical report is not peer reviewed and the preprint has no journal version.",
   "The wet-lab count conflicts within the source (1,000 vs 900).",
   "Weights carry a custom licence and the suite is non-commercial."
  ],
  "sources": [
   {
    "url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf",
    "title": "Technical report",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
    "title": "Preprint version record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://instadeep.com/2026/02/modelling-the-genome-with-ntv3/",
    "title": "Official blog and weights release",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://github.com/instadeepai/nucleotide-transformer",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-12-22 (inferred from the DOI string)",
    "now": "2025-12-25 (preprint v1 posted)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2026-02-18, product update",
    "now": "2026-10-29, preprint v2 (paper revision)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.22.695963",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "600 designs",
    "now": "900 (600+150+150), noting that the source itself conflicts between 1,000 and 900",
    "source_url": "https://instadeep.com/wp-content/uploads/2025/12/NT_v3.pdf",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "wetlab_validation[0]",
   "open_status.code_license",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-ntv3/",
   "en": "/bio/benchmark/models/ntv3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-opencomplex2",
  "kind": "model",
  "short_slug": "opencomplex2",
  "names": {
   "display": "OpenComplex2",
   "full": null,
   "aliases": [
    "OC2",
    "OpenComplex-2"
   ],
   "ambiguous_aliases": [
    "OpenComplex"
   ]
  },
  "family": "OpenComplex",
  "family_null_reason": null,
  "domain": "structure",
  "domains_secondary": [
   "genomics-rna"
  ],
  "one_liner": "All-atom generative model that samples conformational distributions of proteins, nucleic acids and complexes by graph diffusion rather than predicting one structure",
  "org": {
   "names": [
    "Beijing Academy of Artificial Intelligence"
   ],
   "global_entity_ids": []
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Conformational ensemble generation (co-folding with diffusion sampling)",
  "open_status": {
   "code": "unclear",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "A free research trial is open to outside users; neither the paper nor official channels state whether code or weights will be released, and the repository holds only a readme"
  },
  "first_release": {
   "date": "2025-05-02",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1 posted",
   "event_kind": "preprint-posted",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.05.01.651643",
   "verification": "verified",
   "note": "The site recorded 2025-05-01, which is the DOI string date, not a posting date"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-05-02",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.05.01.651643",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-05-02",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.05.01.651643",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-05-03",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v2 posted; no sourced substantive event since",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://www.biorxiv.org/content/10.1101/2025.05.01.651643v2",
   "verification": "verified",
   "description": "The code repository still holds only a readme stating that code will be released, and no journal version was found (as of 2026-11-16)",
   "note": null
  },
  "status": "maintained",
  "status_note": "No new version or code release found within 12 months",
  "what_it_does": "Represents different molecule types as one graph form and runs diffusion sampling at atom, residue and motif levels, shifting the goal from predicting one structure to sampling a conformational distribution, and claims to handle very large assemblies and multiple binding sites. It complements single-structure co-folding such as AlphaFold 3 and Protenix, and unlike BioEmu-2, which also targets dynamics, it additionally claims nucleic-acid and small-molecule coverage.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "The paper describes a purely data-driven system without disclosing dataset composition",
    "source_url": "https://sciety.org/articles/activity/10.1101/2025.05.01.651643"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://sciety.org/articles/activity/10.1101/2025.05.01.651643"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "The abstract mentions incorporating experimental data without stating source or scale",
    "source_url": "https://sciety.org/articles/activity/10.1101/2025.05.01.651643"
   },
   "d_wetlab_validation_only": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://sciety.org/articles/activity/10.1101/2025.05.01.651643"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://predictioncenter.org/casp16/doc/presentations/Day-4/Day4-10-Discussion-QiweiYe-OpenComplex2(WithoutVideo).pdf",
    "confidence": "medium",
    "second_hand": false,
    "note": "It took part in that round and presented on the discussion day; the slides' comparison figures are marked unpublished data and no official rank or score was found"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own ensemble-reproduction claims",
    "category": "own-task-set",
    "detail": "The abstract claims reproduction of known protein and RNA ensembles and capture of allostery and cryptic pockets, with no citable figures",
    "source_type": "self"
   },
   {
    "label": "A general claim of competitive structure prediction",
    "category": "comparison-vs-baseline-only",
    "detail": "The abstract only says competitive, naming no evaluation set or metric",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report producing conformational landscapes within hours across proteins, nucleic acids and small molecules, scaling to very large assemblies.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation was found, and the comparison figures in the participation talk are the developers' own",
  "background": {
   "org_profile": "The developer is a nonprofit research institute founded in Beijing in 2018 as a non-enterprise unit, set up with local science-authority backing; this model belongs to its life-science model series.",
   "funding": "A nonprofit with no funding rounds; the team's budget composition is not public.",
   "products_partnerships": "The previous generation was that series' structure-prediction model; officially the research trial is free and registered users span over three hundred institutions, a registration count rather than usage, and no pharmaceutical partnerships are public.",
   "related_models": [
    "model-bioemu-2",
    "model-alphafold3",
    "model-protenix"
   ]
  },
  "reading_notes": [
   "Its target is a conformational distribution while most existing benchmarks score a single structure, so the protocols do not line up.",
   "Participation is documented by an official talk, but no citable rank or value exists.",
   "The institution count is an official registration figure, not usage and not performance.",
   "Neither code nor weights have a public release statement, so its claim of parity with dynamics simulation cannot be reproduced externally."
  ],
  "caveats": [
   "The site's first-release date used the DOI string; it is corrected to the preprint posting date.",
   "Training data is entirely undisclosed and cannot be filed under any known source class.",
   "The abstract's capability statements carry no figures and should not be restated as evaluation results."
  ],
  "sources": [
   {
    "url": "https://sciety.org/articles/activity/10.1101/2025.05.01.651643",
    "title": "Preprint (aggregator page)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.05.01.651643",
    "title": "Preprint version record",
    "kind": "registry",
    "date": null
   },
   {
    "url": "https://predictioncenter.org/casp16/doc/presentations/Day-4/Day4-10-Discussion-QiweiYe-OpenComplex2(WithoutVideo).pdf",
    "title": "Discussion-day talk at the assessment",
    "kind": "presentation",
    "date": null
   },
   {
    "url": "https://github.com/baaihealth/opencomplex2",
    "title": "Code repository (readme only)",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2025-05-01",
    "now": "2025-05-02 (preprint v1 posted; v2 on 2025-05-03)",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2025.05.01.651643",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.code",
   "open_status.weights",
   "training_data.c_own_wetlab_for_training",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-opencomplex2/",
   "en": "/bio/benchmark/models/opencomplex2/"
  },
  "names_full_null_reason": "The name is a project name; the preprint gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-opendde",
  "kind": "model",
  "short_slug": "opendde",
  "names": {
   "display": "OpenDDE",
   "full": "Open Drug Discovery Engine",
   "full_null_reason": null,
   "aliases": [
    "OpenDDE"
   ],
   "ambiguous_aliases": [
    "DDE"
   ]
  },
  "family": null,
  "family_null_reason": "Covers the open OpenDDE weights only; the company's internal antibody design model is related but different",
  "domain": "structure",
  "domains_secondary": [
   "antibody"
  ],
  "one_liner": "Fully open 655M-parameter co-folding model tuned for antibody-antigen, billed as an open discovery engine",
  "org": {
   "names": [
    "Aureka AI Research",
    "Aureka Biotechnologies"
   ],
   "global_entity_ids": []
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US",
   "CN"
  ],
  "model_class": "Co-folding (antibody-leaning)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code, weights and benchmarks all ship under Apache-2.0, making it a fully reproducible counterpart to closed engines"
  },
  "first_release": {
   "date": "2026-07-04",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv technical report (the HF card notes a 07-03 preview; public announcement 07-06)",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2607.03787",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-07-03",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub preview release (per the HF card)",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/aurekaresearch/OpenDDE",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2026-07-04",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv technical report v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2607.03787",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-07-03",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub preview release",
    "timestamp_utc": null,
    "source_url": "https://github.com/aurekaresearch/OpenDDE/releases",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2026-07-04",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Hugging Face weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/aurekaresearch/OpenDDE",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-07-06",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Public announcement (per press coverage)",
    "timestamp_utc": null,
    "source_url": "https://pharmaphorum.com/news/aureka-offers-open-source-ai-module-drug-discovery",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-08",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release 1.2.0",
   "description": "1.2.0 adds constraint-guided inference with examples, after 1.1.0 (2026-08-16) and several 1.0.x releases",
   "timestamp_utc": null,
   "source_url": "https://github.com/aurekaresearch/OpenDDE/releases",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Shipping releases through its first year",
  "what_it_does": "An all-atom biomolecular foundation model that predicts protein, nucleic-acid and small-molecule complexes as the entry point of a planned discovery engine, tuned for antibody-antigen generalisation. Unlike the proprietary weights of Protenix or the closed IsoDDE, it releases training code, inference pipeline, checkpoints and benchmarks together.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Weighted PDB, AFDB-multimer, distillation data, MGnify, Swiss-Prot and SAbDab",
    "source_url": "https://arxiv.org/abs/2607.03787"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "The paper does not mention private or licensed data",
    "source_url": "https://arxiv.org/abs/2607.03787"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "Not mentioned in the paper; the company has a high-throughput wet lab and calls the feedback loop a later feature",
    "source_url": "https://pharmaphorum.com/news/aureka-offers-open-source-ai-module-drug-discovery"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "self-defined FoldBench antibody-antigen subset",
    "source_url": "https://arxiv.org/abs/2607.03787",
    "confidence": "high",
    "verification": "verified",
    "note": "The antibody subset is self-defined and not comparable with the official board; the appendix gives no threshold for that panel"
   },
   {
    "benchmark_id": "bench-36-aintibody",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": "Challenge 1 (computational affinity maturation)",
    "source_url": "https://www.thailand-business-news.com/pr-news/nature-biotechnology-aureka-wins-the-global-blinded-ai-antibody-benchmark-ai-design-surpasses-the-best-experimental-result",
    "confidence": "high",
    "verification": "unverified",
    "note": "The placing belongs to the company's internal antibody design model, registered under another name in the competition paper, and cannot be credited to these weights; the matrix and the model page still conflict (unverified)"
   }
  ],
  "other_evaluations": [
   {
    "label": "Self-built antibody evaluation sets",
    "category": "self-report",
    "detail": "The report also self-reports on two antibody evaluation sets, one partly self-built, each under both a top-ranked and an oracle selection protocol",
    "source_type": "self"
   },
   {
    "label": "The competition paper's overall finding",
    "category": "third-party-reading",
    "detail": "The AIntibody paper concludes computational design did not beat experimental methods overall; the press release's wording overstates it",
    "source_type": "third"
   },
   {
    "label": "The open-version claim",
    "category": "second-hand",
    "detail": "The claim that this is the open version of the internal model appears only in a company release and one second-hand article, never in the report or repository",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report the first fully open co-folding model at closed-engine level, leading on antibody-antigen tasks (self-reported, on partly self-built sets).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Only self-reported tables exist; no third-party rerun found",
  "background": {
   "org_profile": "Aureka Biotechnologies was founded in 2023, headquartered in Laguna Hills, California with a Shanghai team; an AI-native biotech running a dry-wet closed loop plus its own pipeline, with headcount undisclosed.",
   "funding": "A Series A in 2025-11 (amounts differ across second-hand sources), a $35M Series A+ on 2026-04-28, and a $100M Series B announced 2026-08-10, bringing the total to nearly $200M.",
   "funding_null_reason": null,
   "products_partnerships": "Open-sources this model as a design-engine module, runs its own cardiometabolic, immunology and CNS pipeline, and is building a new wet lab in Irvine; no pharma partnership disclosed.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-isodde",
    "model-protenix",
    "model-esmfold2"
   ]
  },
  "reading_notes": [
   "The antibody figures are high but partly rest on self-built sets and need independent replication.",
   "Each antibody evaluation comes in two selection protocols; any citation must say which one.",
   "Without PoseBusters or Runs N' Poses there is no comparison of small-molecule pose accuracy.",
   "Openness is its firmest advantage: the only fully reproducible engine-level counterpart."
  ],
  "caveats": [
   "The competition placing belongs to the internal model, not these weights; the pairing remains an open inconsistency (unverified).",
   "The open-version framing rests on company wording plus one second-hand article.",
   "The report is internally inconsistent about one baseline model's values.",
   "The company calls the wet-lab feedback loop a later feature; this version uses no in-house data."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2607.03787",
    "title": "OpenDDE technical report",
    "kind": "preprint",
    "date": {
     "date": "2026-07-04",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "arXiv v1",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2607.03787",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://github.com/aurekaresearch/OpenDDE/releases",
    "title": "OpenDDE GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/aurekaresearch/OpenDDE",
    "title": "Hugging Face weights",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41587-026-03238-6",
    "title": "Antibody design competition paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pharmaphorum.com/news/aureka-offers-open-source-ai-module-drug-discovery",
    "title": "Press report on the open module and the company",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.thailand-business-news.com/pr-news/nature-biotechnology-aureka-wins-the-global-blinded-ai-antibody-benchmark-ai-design-surpasses-the-best-experimental-result",
    "title": "Syndicated company press release (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported.bench-36-aintibody",
    "was": "The matrix marked it as an official board entry",
    "now": "Changed to unconfirmed: the placing belongs to the company's internal antibody model, not these weights (unverified)",
    "source_url": "https://www.thailand-business-news.com/pr-news/nature-biotechnology-aureka-wins-the-global-blinded-ai-antibody-benchmark-ai-design-surpasses-the-best-experimental-result",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[1]",
   "background.funding"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-opendde/",
   "en": "/bio/benchmark/models/opendde/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-openfold3",
  "kind": "model",
  "short_slug": "openfold3",
  "names": {
   "display": "OpenFold3",
   "full": null,
   "full_null_reason": "No expanded name; official materials use OpenFold3",
   "aliases": [
    "OpenFold3",
    "OpenFold3-preview",
    "OpenFold3-preview2",
    "OF3"
   ],
   "ambiguous_aliases": [
    "OpenFold"
   ]
  },
  "family": null,
  "family_null_reason": "Single model line; preview, preview2 and 0.5.x are versions of one model",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Consortium-built Apache-2.0 open reproduction of AlphaFold 3 with a federated private-data fine-tuning track",
  "org": {
   "names": [
    "OpenFold Consortium",
    "Open Molecular Software Foundation",
    "Columbia University",
    "Lawrence Livermore National Laboratory",
    "Seoul National University"
   ],
   "global_entity_ids": []
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding (protein / nucleic acid / small molecule)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code and weights Apache-2.0, commercial use allowed; the 0.5.x default weights OpenBind-0 are also Apache-2.0"
  },
  "first_release": {
   "date": "2025-10-28",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub preview release, consortium press release the same day",
   "timestamp_utc": null,
   "source_url": "https://github.com/aqlaboratory/openfold-3/releases",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub preview release",
    "timestamp_utc": null,
    "source_url": "https://github.com/aqlaboratory/openfold-3/releases",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Preview code public",
    "timestamp_utc": null,
    "source_url": "https://github.com/aqlaboratory/openfold-3/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Preview weights public on Hugging Face with the release",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/OpenFold/OpenFold3",
    "verification": "partial",
    "note": "Same-day weight availability not separately checked"
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "Consortium press release",
    "timestamp_utc": null,
    "source_url": "https://www.businesswire.com/news/home/20251028507233/en",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-28",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release 0.5.1",
   "description": "v0.5.1 patch: lower inference/validation memory (MSA/template features as compact indices) and deterministic inference, no new weights; OpenBind-0 weights belong to v0.5.0 (2026-08-21)",
   "timestamp_utc": null,
   "source_url": "https://github.com/aqlaboratory/openfold-3/releases/tag/0.5.1",
   "verification": "verified",
   "note": "Commits continue to 2026-11-16; preview2 was 2026-03-13"
  },
  "status": "active",
  "status_note": "Releases and ongoing commits within 12 months",
  "what_it_does": "Open PyTorch reproduction of AlphaFold 3 for protein, nucleic-acid and small-molecule complexes, with reproducible training and distillation pipelines. Unlike Protenix it adds federated fine-tuning on pharma private data; unlike AlphaFold 3 it differs in licence and reproducibility. AQAffinity is built on it.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "300,000+ experimental structures plus 40M+ synthetic/distilled entries (incl. 13M MGnify)",
    "source_url": "https://github.com/aqlaboratory/openfold-3"
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "Federated fine-tuning by five pharma firms on 20,167 private structures, 1,056 held out (Apheris self-reported)",
    "source_url": "https://www.pharmiweb.com/pwtoday-story/ai-consortium-to-advance-openfold3-for-drug-discovery"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "Whether OpenBind-0 uses newly measured data is not stated",
    "source_url": "https://github.com/aqlaboratory/openfold-3"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer leaderboard (preview)",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "high",
    "verification": "verified",
    "note": "Maintainer leaderboard lists the preview version"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "Runs N' Poses (protein-ligand)",
    "source_url": "https://portal.openfold.omsf.io/reports/of3p2_technical_report.pdf",
    "confidence": "medium",
    "verification": "partial",
    "note": "README bar chart only, no numeric text; methods in the preview2 technical report"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "CASP16 protein and RNA monomers (post hoc)",
    "source_url": "https://github.com/aqlaboratory/openfold-3",
    "confidence": "medium",
    "verification": "partial",
    "note": "CASP16 data used post hoc, README figure only; source type kept as in the pairing table"
   }
  ],
  "other_evaluations": [
   {
    "label": "AF3 antibody-antigen set",
    "category": "public-benchmark-not-on-site",
    "detail": "preview2 protocol, figures only",
    "source_type": "self"
   },
   {
    "label": "Federated fine-tuning holdout",
    "category": "own-task-set",
    "detail": "Evaluation on 1,056 held-out private structures, self-reported by Apheris, not independently verified",
    "source_type": "partner"
   }
  ],
  "claimed_advantage": "Developers report: the preview is the only model matching AlphaFold 3 on CASP16 RNA monomers, under a commercially usable Apache-2.0 licence with reproducible training.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The FoldBench maintainer leaderboard lists the preview, run by the board; federated fine-tuning gains lack independent verification.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "OpenFold is a non-profit AI research consortium hosted by the Open Molecular Software Foundation, with academic and industry members (NVIDIA cites 40+ organizations). OpenFold3 was co-developed by a US university lab, a US national laboratory and a South Korean university group.",
   "funding": "Members contribute funding, compute and expertise; public materials name no specific grantor.",
   "funding_null_reason": null,
   "products_partnerships": "The AISB network, run by Apheris, has AbbVie, Astex, BMS, J&J and Takeda continue training via federated learning (source lists differ slightly); Apheris also offers ApherisFold; NVIDIA ships a NIM package; default weights OpenBind-0 were trained by the UK OpenBind project with a 2025-06 cutoff.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3",
    "model-protenix",
    "model-rf3",
    "model-boltz-1-1x",
    "model-aqaffinity"
   ]
  },
  "reading_notes": [
   "Its most informative feature is the federated route: whether private data that never leaves the owner improves a general model; the only evidence so far is Apheris's own account.",
   "The developers' own evaluations are figures without numbers and cannot be compared directly with other models.",
   "The 2026-10-28 release 0.5.1 is only a patch; new weights belong to 0.5.0 (2026-08-21); do not conflate them.",
   "Apache-2.0 plus reproducible training gives it distinct value where self fine-tuning is needed, independent of leaderboard rank."
  ],
  "caveats": [
   "No PoseBusters result; it is not in the official evaluation set.",
   "Sources disagree slightly on the list of federated pharma partners.",
   "The CASP16 pairing is post-hoc dataset reuse, not participation."
  ],
  "sources": [
   {
    "url": "https://github.com/aqlaboratory/openfold-3",
    "title": "OpenFold3 GitHub repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/aqlaboratory/openfold-3/releases",
    "title": "OpenFold3 GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20251028507233/en",
    "title": "OpenFold Consortium press release (2025-10-28)",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://portal.openfold.omsf.io/reports/of3p2_technical_report.pdf",
    "title": "OpenFold3-preview2 technical report",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/OpenFold/OpenFold3",
    "title": "OpenFold3 on Hugging Face",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://build.nvidia.com/openfold/openfold3/modelcard",
    "title": "NVIDIA OpenFold3 model card",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.pharmiweb.com/pwtoday-story/ai-consortium-to-advance-openfold3-for-drug-discovery",
    "title": "AISB federated fine-tuning report",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.tamarind.bio/tools/openbind",
    "title": "OpenBind third-party platform page (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-08-21",
    "now": "2026-10-28 (tag 0.5.1)",
    "source_url": "https://github.com/aqlaboratory/openfold-3/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "v0.5.0 OpenBind model release: OpenBind-0 weights and recipes",
    "now": "v0.5.1 patch: memory reduction and deterministic inference, no new weights; OpenBind-0 weights belong to v0.5.0 (2026-08-21)",
    "source_url": "https://github.com/aqlaboratory/openfold-3/releases/tag/0.5.1",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.event_kind",
    "was": "new-weights",
    "now": "code-release",
    "source_url": "https://github.com/aqlaboratory/openfold-3/releases/tag/0.5.1",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.weights_release",
   "training_data.b_private_licensed"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-openfold3/",
   "en": "/bio/benchmark/models/openfold3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-origin-1",
  "kind": "model",
  "short_slug": "origin-1",
  "names": {
   "display": "Origin-1",
   "full": null,
   "aliases": [
    "Origin-1",
    "AbsciGen",
    "AbsciBind"
   ],
   "ambiguous_aliases": [
    "Origin"
   ]
  },
  "family": null,
  "family_null_reason": "A single platform whose components are described inside one record",
  "domain": "antibody",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Generative platform designing full-length antibodies de novo against zero-prior epitopes, combining diffusion generation, paired CDR sequence design and co-folding scoring",
  "org": {
   "names": [
    "Absci"
   ],
   "global_entity_ids": [
    "absci"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (all-atom diffusion, sequence design, scoring)",
  "open_status": {
   "code": "partial",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Data and evaluation are in a public repository; neither the paper nor the website states whether weights are released or under what terms"
  },
  "first_release": {
   "date": "2026-01-14",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1 posted, with a developer launch page the same day",
   "event_kind": "preprint-posted",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.01.14.699389",
   "verification": "verified",
   "note": "Materials giving 2026-03 refer to a results-disclosure date, not first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-01-14",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.01.14.699389",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-01-14",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.01.14.699389",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-01-14",
    "precision": "day",
    "year_hint": null,
    "basis": "Data and evaluation scripts in a public repository; weights not included",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/AbSciBio/origin-1",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-01-14",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer launch page",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://www.absci.com/denovo/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-05-05",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v2",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.01.14.699389",
   "verification": "verified",
   "description": "The preprint was revised to v2; afterwards the public repository saw only maintenance edits to data tables and the readme (latest 2026-11-13), and no journal version was found",
   "note": null
  },
  "status": "active",
  "status_note": "Preprint revision and repository maintenance within 12 months",
  "what_it_does": "Conditions on an unbound antigen structure, an antibody framework and a user-specified epitope to generate the complex by all-atom diffusion, designs paired heavy and light chain sequences for six CDRs, then screens with co-folding scoring before the wet lab. Against RFantibody it is all-atom, allows antigen flexibility and explicitly targets zero-prior epitopes; against BoltzGen its antibody evidence lies in full-length antibodies rather than nanobodies.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "An in-house antibody structure set from 10,045 public structure-database entries, covering all entries of a dedicated antibody database, with homologous complexes of the zero-prior targets excluded from every split",
    "source_url": "https://www.absci.com/wp-content/uploads/2026/01/260114_Origin1_Final_Submission.pdf"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://www.absci.com/wp-content/uploads/2026/01/260114_Origin1_Final_Submission.pdf"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "The methods section describes only public structure data, while the website also mentions years of proprietary data generation and closed-loop experiments without scale or composition",
    "source_url": "https://www.absci.com/denovo/"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Binding, specificity, aggregation, structural and functional assays across ten targets",
    "source_url": "https://www.absci.com/denovo/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Zero-prior epitope design (ten targets)",
    "targets": 10,
    "designs_tested": null,
    "hits": 4,
    "denominator_basis": "Ten targets with fewer than 100 designs each; hits counted per target as developable antibodies passing several orthogonal assays",
    "success_rate": "4/10 targets",
    "affinity_best": null,
    "version": "preprint v1 / developer page",
    "source_type": "self",
    "source_url": "https://www.absci.com/denovo/",
    "date": {
     "date": "2026-01-14",
     "precision": "day",
     "year_hint": null,
     "basis": "Preprint and launch page, same day",
     "event_kind": "preprint-posted",
     "timestamp_utc": null,
     "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.01.14.699389",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Run by the developer; no independent experimental arm",
    "unverified_marker": null,
    "note": "Two of the targets also have cryo-EM structures matching the designs at 3.0 A and 3.1 A; the developers note few hits per target, few designs and stricter acceptance criteria"
   },
   {
    "campaign": "Affinity maturation on one target",
    "targets": 1,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "An AI-guided single-mutation library whose size and number tested are not given on the launch page",
    "success_rate": null,
    "affinity_best": null,
    "version": "developer page",
    "source_type": "self",
    "source_url": "https://www.absci.com/denovo/",
    "date": {
     "date": "2026-01-14",
     "precision": "day",
     "year_hint": null,
     "basis": "Launch page",
     "event_kind": "official-announcement",
     "timestamp_utc": null,
     "source_url": "https://www.absci.com/denovo/",
     "verification": "partial",
     "note": null
    },
    "verification": "partial",
    "executor": "Run by the developer",
    "unverified_marker": "The exact potency previously published is not on the launch page, which gives only an approximate value; it must be sourced from the paper before being shown",
    "note": "The developers report a functional antagonist after maturation with a tens-fold affinity gain; self-reported"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "In-silico comparison with an open antibody design model",
    "category": "comparison-vs-baseline-only",
    "detail": "Compares the share of designs passing the developer's own score on four zero-prior targets; not an experimental hit-rate comparison",
    "source_type": "self"
   },
   {
    "label": "Own assessment of the scoring module",
    "category": "own-task-set",
    "detail": "Separates binders from non-binders across eight antibody-antigen systems against several co-folding models, with some comparator values taken from other papers",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report de novo full-length antibodies against targets with no known complex structure, with four of ten targets yielding developable binders from fewer than 100 designs each.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party assessment or replication was found; earlier critiques of this developer concern a separate 2023 piece of work, not this model",
  "background": {
   "org_profile": "The developer is a US-listed AI-native drug discovery company headquartered in Washington state with sites in New York, Switzerland and Serbia, and 140 employees at end-2025 per its annual report.",
   "funding": "A listed company; in 2025-01 a semiconductor company subscribed $20M of its shares per the annual report.",
   "products_partnerships": "The annual report lists collaborations with several research institutions and companies; the clinical pipeline centres on one antibody candidate and no trial application tied to this model is disclosed.",
   "related_models": [
    "model-rfantibody",
    "model-boltzgen",
    "model-chai-1"
   ]
  },
  "reading_notes": [
   "It is relatively strong public evidence for zero-prior epitopes with full-length antibodies and structural confirmation, but hits per target and design counts are small.",
   "The gap against the open model uses the developer's own score on only four targets and is not an experimental hit rate.",
   "No antibody benchmark on this site lists it, so generalisation can only be judged by third parties reproducing it on zero-prior targets.",
   "Whether weights are public is unstated, which limits external reproduction."
  ],
  "caveats": [
   "All results are self-reported, with no peer-reviewed version and no independent experimental arm.",
   "The exact affinity-maturation value is absent from the launch page and must be sourced from the paper before citation.",
   "The proprietary data and closed-loop experiments mentioned on the website come without scale, so training data composition cannot be judged from them.",
   "Under this round's relation rules it has no countable pairing with any benchmark on this site."
  ],
  "sources": [
   {
    "url": "https://www.absci.com/wp-content/uploads/2026/01/260114_Origin1_Final_Submission.pdf",
    "title": "Preprint (developer PDF)",
    "kind": "preprint",
    "date": "2026-01-14"
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.64898/2026.01.14.699389",
    "title": "Preprint version record",
    "kind": "registry",
    "date": null
   },
   {
    "url": "https://www.absci.com/denovo/",
    "title": "Developer launch page",
    "kind": "official-page",
    "date": null
   },
   {
    "url": "https://github.com/AbSciBio/origin-1",
    "title": "Data and evaluation repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "Affinity maturation gives exact potency values",
    "now": "The release page gives only approximate values; exact values must be taken from the paper, so none are shown for now",
    "source_url": "https://www.absci.com/denovo/",
    "date_found": "2026-11-16"
   },
   {
    "field": "training_data.c_own_wetlab_for_training",
    "was": "Own experimental data used in training",
    "now": "The paper's methods section mentions only public structure data and the website claim has no scale, so it is treated as not disclosed",
    "source_url": "https://www.absci.com/denovo/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.weights",
   "training_data.c_own_wetlab_for_training",
   "wetlab_validation[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-origin-1/",
   "en": "/bio/benchmark/models/origin-1/"
  },
  "names_full_null_reason": "The name is a platform name; the preprint gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No model page in the site-wide entity library; only the developer organisation has one",
  "status_head": null
 },
 {
  "id": "model-pearl",
  "kind": "model",
  "short_slug": "pearl",
  "names": {
   "display": "Pearl",
   "full": null,
   "full_null_reason": "No expanded name; official materials use Pearl",
   "aliases": [
    "Pearl",
    "Pearl-2026.1.dev",
    "Pearl system",
    "DeCAF-Pearl"
   ],
   "ambiguous_aliases": [
    "Pearl"
   ]
  },
  "family": null,
  "family_null_reason": "Single model line; Pearl system and DeCAF-Pearl are its inference system and distilled variant",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Genesis Molecular AI's closed protein-ligand co-folding model, trained on large-scale physics-based synthetic structures",
  "org": {
   "names": [
    "Genesis Molecular AI"
   ],
   "global_entity_ids": [
    "genesis-molecular-ai"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding (protein-ligand, SO(3)-equivariant diffusion)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Neither the paper nor the release page offers weights or inference code; no API found"
  },
  "first_release": {
   "date": "2025-10-28",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv v1, with the Genesis release post the same day",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2510.24670",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2510.24670",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv 2510.24670 v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2510.24670",
    "verification": "verified",
    "note": "v2 followed the next day; evaluation protocol is stated in the v2 text"
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-10-28",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Genesis release post, with the same-day company rename press release",
    "timestamp_utc": null,
    "source_url": "https://www.genesis.ml/news/introducing-pearl",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-06-10",
   "precision": "day",
   "event_kind": "product-update",
   "basis": "Genesis blog post: DeCAF-Pearl distillation",
   "description": "Announced DeCAF-Pearl (flow-map distillation on the Pearl-2026.1.dev checkpoint), after the 06-03 Pearl system with inference-time scaling and pose ranking",
   "timestamp_utc": null,
   "source_url": "https://www.genesis.ml/news/genesis-model-distillation",
   "verification": "verified",
   "note": "Blog page dates taken from search results; the model remains proprietary, with no weights or API found"
  },
  "status": "active",
  "status_note": "Official product updates within 12 months; the model itself is closed",
  "what_it_does": "Predicts protein-ligand complex structures, with pocket conditioning and templates at inference. The paper says performance scales with synthetic data, and concedes out-of-distribution accuracy drops and confidence-based pose selection is unreliable. Unlike Boltz-2 it targets pose accuracy, not affinity; unlike AlphaFold 3 it relies on synthetic data.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB (to 2021-09-30) plus OpenFold / AFDB monomer distillation, and physics-generated synthetic structures",
    "source_url": "https://arxiv.org/html/2510.24670v2"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The paper says no proprietary experimental data was used",
    "source_url": "https://arxiv.org/html/2510.24670v2"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "Synthetic structures are physics-generated (582,065 in the scaling experiment), not wet lab; flagship set size undisclosed",
    "source_url": "https://arxiv.org/html/2510.24670v2"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "111 existing internal crystal structures used for evaluation only, not new wet lab",
    "source_url": "https://arxiv.org/html/2510.24670v2"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Internal crystal-structure holdout (existing data)",
    "modality": "structure",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "111 internal crystal structures used for evaluation only; training strictly on public data",
    "success_rate": null,
    "affinity_best": null,
    "version": "arXiv v2",
    "source_type": "self",
    "source_url": "https://arxiv.org/html/2510.24670v2",
    "date": null,
    "verification": "verified",
    "note": "Developer-reported; the set is private and cannot be reproduced externally, not an independent replication; the synthetic set is not wet lab"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "PoseBusters V2, best@5 with PB-valid filter",
    "source_url": "https://arxiv.org/html/2510.24670v2",
    "confidence": "medium",
    "verification": "partial",
    "note": "Self-reported best@5 with PB-valid; AlphaFold 3 taken from its official max-confidence figure, a mixed protocol; Boltz-2 excluded for its later training cutoff"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "Runs N' Poses, best@5 with PB-valid filter",
    "source_url": "https://arxiv.org/html/2510.24670v2",
    "confidence": "medium",
    "verification": "partial",
    "note": "Self-reported in the paper; baselines were re-run by the Pearl team"
   }
  ],
  "other_evaluations": [
   {
    "label": "InternalXtals internal crystal set",
    "category": "own-task-set",
    "detail": "111 internal crystal structures, unconditioned and pocket-conditioned; private and not reproducible",
    "source_type": "self"
   },
   {
    "label": "OpenBind zero-shot evaluation (Pearl system)",
    "category": "public-benchmark-not-on-site",
    "detail": "Company blog says Pearl system beats all co-folding models zero-shot; company account only",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: it beats AlphaFold 3 and open models on PoseBusters and Runs N' Poses (best@5 with PB-valid), with performance scaling with synthetic data.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party replication; weights unreleased",
  "background": {
   "org_profile": "Genesis Molecular AI (formerly Genesis Therapeutics), founded in 2019, is based in Burlingame, California, with its own lab in San Diego and an office in New York; an AI-native small-molecule drug discovery company, it renamed itself and released Pearl on 2025-10-28.",
   "funding": "$52M Series A (2020-12); $200M Series B (2023-08, a16z co-led, NVentures among participants); further NVentures equity investment in 2024-11. A cumulative figure above $280M is second-hand only.",
   "funding_null_reason": null,
   "products_partnerships": "Pearl is the structure engine of the GEMS platform; NVIDIA engineering work speeds training and inference (company figures); pharma partners include Gilead and Incyte, whose 2026-05-20 expansion brings $120M upfront and proprietary experimental data to train Genesis's next models.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3",
    "model-boltz-2",
    "model-boltz-1-1x",
    "model-chai-1",
    "model-protenix"
   ]
  },
  "reading_notes": [
   "Both headline results are author-reported under best@5 with PB-valid; with single predictions or without PB-valid, both numbers and ranking change.",
   "The comparison takes AlphaFold 3 at its official max-confidence protocol, below other papers' re-runs, which inflates the margin.",
   "The synthetic-data route is notable, but the paper admits out-of-distribution decline, matching the Runs N' Poses pattern of lower similarity, worse results.",
   "Gains on the internal crystal set cannot be reproduced externally and are company claims only."
  ],
  "caveats": [
   "Other models' results in the paper were re-run or cited by the Pearl team; this page does not reproduce them.",
   "Neither weights nor inference code are public, so third parties cannot check the results.",
   "Pearl is a common word; searches and auto-linking can hit unrelated projects of the same name."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2510.24670",
    "title": "Pearl paper (arXiv 2510.24670)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2510.24670v2",
    "title": "Pearl paper v2 full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.genesis.ml/news/introducing-pearl",
    "title": "Genesis: Introducing Pearl",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://genesis.ml/wp-content/uploads/2025/10/pearl_technical_report-1-1.pdf",
    "title": "Pearl technical report PDF",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.genesis.ml/news/genesis-model-distillation",
    "title": "Genesis: DeCAF-Pearl model distillation",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.genesis.ml/news/zero-shot-pearl-system-surpasses-all-cofolding-models-on-openbind",
    "title": "Genesis: Pearl system zero-shot on OpenBind",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20251028030745/en/Genesis-Molecular-AI-Unveils-Pearl-a-Field-Leading-Foundation-Model-that-Achieves-Unprecedented-Performance-in-Drug-Protein-Structure-Prediction",
    "title": "Genesis Molecular AI Pearl launch press release (2025-10-28)",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20230821419353/en/Genesis-Therapeutics-Closes-Oversubscribed-$200-Million-Series-B",
    "title": "Genesis Series B announcement",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://techcrunch.com/2020/12/02/genesis-therapeutics-raises-52m-a-round-for-its-ai-focused-drug-discovery-mission/",
    "title": "TechCrunch: Genesis Series A",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20241113952492/en/Genesis-Therapeutics-Announces-New-Equity-Investment-and-Collaboration-with-NVIDIA-to-Optimize-Key-Computational-Methods-for-AI-powered-Drug-Discovery",
    "title": "Genesis NVIDIA investment and collaboration announcement",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://investor.incyte.com/news-releases/news-release-details/incyte-and-genesis-expand-molecular-ai-collaboration-accelerate",
    "title": "Incyte and Genesis collaboration expansion press release",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-pearl/",
   "en": "/bio/benchmark/models/pearl/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-pinal",
  "kind": "model",
  "short_slug": "pinal",
  "names": {
   "display": "Pinal",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "Pinal",
    "Denovo-Pinal"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "protein",
  "domains_secondary": [],
  "one_liner": "Natural-language-to-protein design: turns a function description into structure tokens, then a sequence",
  "org": {
   "names": [
    "Westlake University"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Text-to-protein generation",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Weights and code on GitHub and Hugging Face under Apache-2.0; the full training corpus is withheld"
  },
  "first_release": {
   "date": "2024-08-02",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 preprint",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.08.01.606258",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-08-02",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.08.01.606258",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-10-21",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Nature online publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41586-026-11205-z",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-10-21",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Nature online publication",
   "description": "Nature online publication; later GitHub and Hugging Face updates are minor documentation changes",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1038/s41586-026-11205-z",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication within 12 months",
  "what_it_does": "Translates a natural-language function description into structure tokens and then designs a sequence; about 16B parameters trained on 1.7B protein-text pairs. Built on the same team's SaProt.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "1.7B synthetic protein-text pairs generated from public sequence, structure, domain and enzyme annotation resources",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Wet lab validated four tasks only",
    "source_url": null
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Fluorescent protein",
    "modality": "protein",
    "targets": null,
    "designs_tested": 29,
    "hits": 1,
    "denominator_basis": "50,000 sequences generated, 29 filtered candidates tested, 1 fluoresced green",
    "success_rate": "1 fluorescent design",
    "affinity_best": null,
    "version": "bioRxiv v9",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.08.01.606258v9.full",
    "date": null,
    "verification": "verified",
    "note": "Verified against the v9 full text, so the earlier unverifiable flag is lifted; developer-run"
   },
   {
    "campaign": "Metabolic H protein",
    "modality": "protein",
    "targets": null,
    "designs_tested": 20,
    "hits": 5,
    "denominator_basis": "50,000 initial candidates filtered to ~1,000 and 20 synthesised: 14/20 soluble, 5 active in a reconstituted system",
    "success_rate": "14/20 (70%) soluble; 5 active in a reconstituted CO2-fixation system",
    "affinity_best": "glycine titer 11.1 mM, 1.7x the native protein",
    "version": "bioRxiv v9",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.08.01.606258v9.full",
    "date": null,
    "verification": "verified",
    "note": "Verified against the v9 full text; developer-run"
   },
   {
    "campaign": "PET hydrolase",
    "modality": "enzyme",
    "targets": null,
    "designs_tested": 94,
    "hits": null,
    "denominator_basis": "200,000 generated, 2,416 passed filtering, 94 synthesised and screened; no hit rate reported",
    "success_rate": null,
    "affinity_best": "best variant Tm 56.84 ± 0.29 C",
    "version": "bioRxiv v9",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.08.01.606258v9.full",
    "date": null,
    "verification": "verified",
    "note": "Previously missing counts are now filled in; developer-run"
   },
   {
    "campaign": "Alcohol dehydrogenase",
    "modality": "enzyme",
    "targets": null,
    "designs_tested": 8,
    "hits": 5,
    "denominator_basis": "2,500 initial candidates, 1,000 high-confidence, 8 selected, 5 expressed and purified across three subfamilies",
    "success_rate": "5 expressed and purified across three subfamilies",
    "affinity_best": null,
    "version": "bioRxiv v9",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.08.01.606258v9.full",
    "date": null,
    "verification": "verified",
    "note": "Substrate conversion shown chromatographically against a no-enzyme control with no numeric activity; developer-run"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Computational evaluation",
    "category": "own-task-set",
    "detail": "Foldability, description-function alignment and sequence diversity versus baselines; no summary number",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report natural-language descriptions can be turned into functional proteins, with activity seen in four tasks without iterative optimization",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "Academic team at Westlake University with collaborators at several universities, institutes and one biotech company; not a company",
   "funding": "Funded by Chinese national and provincial research grants",
   "funding_null_reason": null,
   "products_partnerships": "Built on the same team's SaProt; the full corpus is withheld for dual-use reasons",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-saprot",
    "model-progen3"
   ]
  },
  "reading_notes": [
   "All four wet-lab tasks are developer-run at a scale of tens of designs",
   "All counts are verified in the v9 full text, so figures can be given"
  ],
  "caveats": [
   "No results on any of the 44 benchmarks; on the zero-pairing list under this round's rules",
   "The full training corpus is withheld, so external reproduction is incomplete"
  ],
  "sources": [
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.08.01.606258",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1038/s41586-026-11205-z",
    "title": "Crossref record 10.1038/s41586-026-11205-z",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/westlake-repl/Denovo-Pinal",
    "title": "GitHub repository westlake-repl/Denovo-Pinal",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/westlake-repl/Pinal",
    "title": "Hugging Face page westlake-repl/Pinal",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41586-026-11205-z",
    "title": "Journal article DOI 10.1038/s41586-026-11205-z",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2024.08.01.606258v9",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-pinal/",
   "en": "/bio/benchmark/models/pinal/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-poet-2",
  "kind": "model",
  "short_slug": "poet-2",
  "names": {
   "display": "PoET-2",
   "full": null,
   "full_null_reason": "The paper uses the model name only, with no expanded form",
   "aliases": [
    "PoET-2"
   ],
   "ambiguous_aliases": [
    "PoET"
   ]
  },
  "family": null,
  "family_null_reason": "The earlier PoET is a separate model; this record covers PoET-2 only",
  "domain": "protein",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "182M-parameter multimodal retrieval-augmented protein foundation model using family context",
  "org": {
   "names": [
    "OpenProtein.AI"
   ],
   "global_entity_ids": []
  },
  "org_type": "platform-company",
  "org_type_note": null,
  "country": [
   "SG"
  ],
  "model_class": "Protein language model (in-context learning)",
  "open_status": {
   "code": "partial",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The abstract says weights are public in a repository, but neither repository nor terms were verified this round; it is also sold as a commercial platform (unverified)"
  },
  "first_release": {
   "date": "2025-08-05",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "Preprint v1 submission",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2508.04724",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-08-05",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2508.04724",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-08-05",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2508.04724",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-08-05",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "The company publications page carries the work from the same period",
    "timestamp_utc": null,
    "source_url": "https://openprotein.ai/publications/multimodal-foundation-model-controllable-protein-generation/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-02-26",
   "precision": "day",
   "event_kind": "paper-revision",
   "basis": "Preprint v2 revision",
   "description": "Paper revised to v2 (v1 was 2025-08-05); no new weights or product tier found",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2508.04724v2",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Paper revised within 12 months and the platform is still sold",
  "what_it_does": "Adds multimodal inputs such as structure on top of its predecessor and uses a set of homologous sequences as context to score or generate new sequences, pitched as a small model needing little data. It shares the retrieval-augmented route with Profluent-E1, which uses its predecessor as a comparison, and resembles VenusREM but is built around a platform product.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "62 million homolog sets from a public sequence database, with public predicted structures for structure conditioning rather than experimental ones",
    "source_url": "https://arxiv.org/abs/2508.04724"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed training data reported",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "zero-shot (substitutions, indels, higher-order, clinical) plus supervised settings",
    "source_url": "https://arxiv.org/html/2508.04724v1",
    "confidence": "high",
    "verification": "verified",
    "note": "Developer-reported in a paper table under the official protocol, not separately confirmed on the official board (self-reported)"
   }
  ],
  "other_evaluations": [
   {
    "label": "Vendor low-data claim",
    "category": "comparison-vs-baseline-only",
    "detail": "Marketing material claims about 30x less experimental data with no public protocol, a vendor figure (unverified)",
    "source_type": "self"
   },
   {
    "label": "Ensemble with another model",
    "category": "own-task-set",
    "detail": "The paper also reports an ensemble setting with a retrieval-augmented model, a developer-built control",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that multimodal retrieval augmentation and family in-context learning let a 182M model lead zero-shot tasks, especially indels and higher-order mutants, and claim about 30x less experimental data (a vendor figure, not an independent evaluation).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Self-reported results are not confirmed on the official board or by third parties",
  "background": {
   "org_profile": "A platform company headquartered in a Singapore biomedical park and listed in a commercial database as founded in 2021. It offers a no-code protein design platform, free for universities and licensed to pharma, with no public headcount.",
   "funding": "A commercial database lists two funds as investors with an early venture round logged in 2024-01 and no amount disclosed (second-hand); the company was also selected for a government research program on next-generation protein foundation models.",
   "funding_null_reason": null,
   "products_partnerships": "The earlier PoET was a 2023 conference paper and PoET-2 followed in 2025. A large pharma company has used the platform since 2025 and announced an expansion into antibody discovery on 2026-03-31, with terms undisclosed. Competitors include Profluent-E1.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-profluent-e1",
    "model-venusrem",
    "model-prosst"
   ]
  },
  "reading_notes": [
   "The self-reported results are not separately confirmed on the official board and there is no public submission record.",
   "Scores on the same benchmark are not directly comparable between retrieval-augmented and sequence-only models.",
   "Reaching the top tier at 182M parameters shows parameters are not the only factor, but that is not a cross-family conclusion.",
   "The vendor's \"30x less data\" is marketing, not an independent evaluation."
  ],
  "caveats": [
   "Openness unverified: the abstract claims public weights, but neither repository nor licence terms were checked this round.",
   "No wet-lab validation; all evaluation is computational.",
   "A commercial platform and open weights are claimed side by side and must be cited separately."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2508.04724",
    "title": "PoET-2 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2508.04724v1",
    "title": "Preprint v1 full text with the self-reported table",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://openprotein.ai/publications/multimodal-foundation-model-controllable-protein-generation/",
    "title": "Company publications page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://news.mit.edu/2026/bringing-ai-driven-protein-design-tools-everywhere-0417",
    "title": "University news story on the platform",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://pitchbook.com/profiles/company/527579-74",
    "title": "Commercial database funding entry (second-hand)",
    "kind": "second-hand",
    "date": null
   },
   {
    "url": "https://www.openprotein.ai/openprotein-ai-selected-for-darpa-program",
    "title": "Company announcement: selected for a government program",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym",
    "title": "ProteinGym official repository",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.event_kind",
    "was": "Recorded as a new version",
    "now": "Changed to a paper revision: 2026-02-26 is preprint v2 and the description itself says there are no new weights",
    "source_url": "https://arxiv.org/abs/2508.04724",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "Said only \"v2 revision, no new weights or product tier\" without tying the dates together",
    "now": "Adds the v1 and v2 dates alongside the unchanged \"no new weights\" statement",
    "source_url": "https://arxiv.org/abs/2508.04724",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.code",
   "open_status.weights",
   "open_status.weights_license",
   "open_status.code_license",
   "background.funding",
   "other_evaluations[0]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-poet-2/",
   "en": "/bio/benchmark/models/poet-2/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-popeve",
  "kind": "model",
  "short_slug": "popeve",
  "names": {
   "display": "popEVE",
   "full": null,
   "aliases": [
    "popEVE"
   ],
   "ambiguous_aliases": [
    "EVE"
   ]
  },
  "family": null,
  "family_null_reason": "The predecessor EVE has no entity record, so no family entry",
  "domain": "genomics-rna",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Proteome-wide missense pathogenicity model with scores comparable across genes",
  "org": {
   "names": [
    "Harvard Medical School (Blavatnik Institute)",
    "Centre for Genomic Regulation (CRG), Barcelona"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US",
   "ES"
  ],
  "model_class": "Missense pathogenicity prediction (evolution + population constraint)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": "GPL-3.0",
   "license_changed": null,
   "note": "Code is open with an online per-gene portal; whether weights are released separately is unclear"
  },
  "first_release": {
   "date": "2023-11-28",
   "precision": "day",
   "year_hint": null,
   "basis": "medRxiv v1 (10.1101/2023.11.27.23299062)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://www.medrxiv.org/content/10.1101/2023.11.27.23299062",
   "verification": "verified",
   "note": "The site said only \"preprint 2023\"; v1 carried a different title from the retitled v2 (2025-03)"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-11-28",
    "precision": "day",
    "year_hint": null,
    "basis": "medRxiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.medrxiv.org/content/10.1101/2023.11.27.23299062",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2023-11-28",
    "precision": "day",
    "year_hint": null,
    "basis": "medRxiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.medrxiv.org/content/10.1101/2023.11.27.23299062",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-11-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal online (received 2025-02-26, accepted 2025-10-08)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41588-025-02400-1",
    "verification": "partial",
    "note": "The online date comes from an institutional press release, not the article header"
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-11-24",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41588-025-02400-1",
   "verification": "partial",
   "description": "Journal publication (DOI 10.1038/s41588-025-02400-1); medRxiv v2 retitled 2025-03; last code commit 2024-07-01",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication within 12 months; no code commits since 2024-07",
  "what_it_does": "Adds a protein language model and population constraint (a public allele-frequency resource and a large cohort) on top of an evolutionary variational model, then calibrates scores proteome-wide with a Gaussian process so variants in different genes sit on one scale - built for ranking which variant in a patient genome is most likely causal. Like AlphaMissense it predicts missense effects, but it emphasises cross-gene calibration and rare-disease cohorts rather than a precomputed proteome-wide catalogue.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Evolutionary MSAs, a public allele-frequency resource, constraint from a large population cohort, and public DMS data",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12695638/"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12695638/"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No own wet lab; new candidate genes were confirmed by external studies",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12695638/"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-11-clinvar",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12695638/",
    "confidence": "high",
    "second_hand": false,
    "note": "Expert-curated 2019 and 2020 subsets (at least five benign and five pathogenic per gene), metric AUC; per-model values appear only in extended data figures"
   }
  ],
  "other_evaluations": [
   {
    "label": "Severe developmental disorder cohort",
    "category": "own-task-set",
    "detail": "More than 31,000 families; the developers report the causal variant ranked first in most cases, hundreds of genes and over a hundred new candidates, with some associations reported as independently confirmed in press coverage",
    "source_type": "self"
   },
   {
    "label": "Cross-gene comparability assessment",
    "category": "own-task-set",
    "detail": "The paper argues the public held-out protocol lacks resolution between leading models, citing rank changes between the two years",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: once scores are calibrated proteome-wide, variants in different genes can be ranked directly, putting the causal variant first in most cases of a large developmental-disorder cohort.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation on a site benchmark was found; external confirmation of candidate genes concerns gene associations, not model evaluation",
  "background": {
   "org_profile": "An academic project: an institute at a US medical school working with a genomics research centre in Barcelona on evolutionary sequence models and variant-effect prediction, with a UK genome institute as a partner. Not a company project.",
   "funding": "Funding from a technology philanthropy's neurodegeneration network, awards from two US federal science agencies and a Catalan government research-centre programme (per acknowledgements and press release; the list may be incomplete).",
   "products_partnerships": "Builds on the group's earlier evolutionary variational model; scores are being integrated into two public variant-annotation resources, with collaborations covering two children's-hospital rare-disease networks and a national genomics programme. Clinical deployment still needs safety and accuracy validation.",
   "related_models": [
    "model-alphamissense"
   ]
  },
  "reading_notes": [
   "On the ClinVar protocol only a qualitative \"comparable\" claim appears and the text gives no AUC, so it cannot be read as beating AlphaMissense.",
   "Variant labels and predictors can contaminate each other, so the protocol's numbers need gene-level assessment alongside.",
   "The cohort results are closer to real use, but gene counts and hit definitions appear in several framings, so always state the denominator.",
   "CAGI does not appear in the full text, so the old candidate pairing does not hold."
  ],
  "caveats": [
   "Per-model AUC values sit only in extended data figures, not the main text.",
   "The cohort gene count has two framings and press coverage differs from the paper.",
   "The journal online date comes from a press release, not the article header; partly verified.",
   "No code commits since 2024-07, and clinical deployment still needs validation."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41588-025-02400-1",
    "title": "Journal paper 2025",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12695638/",
    "title": "Journal full text (public archive)",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.medrxiv.org/content/10.1101/2023.11.27.23299062",
    "title": "Preprint v1",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/debbiemarkslab/popEVE",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://hms.harvard.edu/news/new-artificial-intelligence-model-could-speed-rare-disease-diagnosis",
    "title": "Institutional press release",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "Preprint 2023 (no date)",
    "now": "2023-11-28（medRxiv v1）",
    "source_url": "https://www.medrxiv.org/content/10.1101/2023.11.27.23299062",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Candidate pairing CAGI",
    "now": "Not a pairing: the benchmark does not appear in the full text",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12695638/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "release_milestones.journal_online",
   "open_status.weights"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-popeve/",
   "en": "/bio/benchmark/models/popeve/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-profluent-e1",
  "kind": "model",
  "short_slug": "profluent-e1",
  "names": {
   "display": "Profluent-E1",
   "full": null,
   "aliases": [
    "Profluent-E1",
    "E1-600m"
   ],
   "ambiguous_aliases": [
    "E1"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "E1-150M",
     "id_or_null": null,
     "first_release": {
      "date": "2025-11-13",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint and weights released together",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://huggingface.co/Profluent-Bio/E1-600m",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "E1-300M",
     "id_or_null": null,
     "first_release": {
      "date": "2025-11-13",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint and weights released together",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://huggingface.co/Profluent-Bio/E1-600m",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "E1-600M",
     "id_or_null": null,
     "first_release": {
      "date": "2025-11-13",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint and weights released together",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://huggingface.co/Profluent-Bio/E1-600m",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "Three parameter sizes from one paper sharing training set and licence; the sibling generative model is excluded"
  },
  "family_null_reason": null,
  "domain": "protein",
  "domains_secondary": [],
  "one_liner": "Retrieval-augmented protein encoder for mutation-effect scoring and representation",
  "org": {
   "names": [
    "Profluent"
   ],
   "global_entity_ids": [
    "profluent"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Protein language model (retrieval-augmented encoder)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code under Apache-2.0; weights under a permissive licence with attribution, which the developer states is free for research and commercial use"
  },
  "first_release": {
   "date": "2025-11-13",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1, with weights posted at the same time",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2025.11.12.688125",
   "verification": "unverified",
   "note": "The DOI request failed, so the date was not reverified this round"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-11-09",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights repository posted",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/Profluent-Bio/E1-600m",
    "verification": "unverified",
    "note": "Repository dates may not serve as first-release evidence and are recorded only"
   },
   "paper_v1": {
    "date": "2025-11-13",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.11.12.688125",
    "verification": "unverified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-11-13",
    "precision": "day",
    "year_hint": null,
    "basis": "Code repository made public",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/Profluent-AI/E1",
    "verification": "unverified",
    "note": null
   },
   "weights_release": {
    "date": "2025-11-13",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights for three sizes made public",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/Profluent-Bio/E1-600m",
    "verification": "unverified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-03-09",
   "precision": "day",
   "year_hint": null,
   "basis": "Repository changelog entry",
   "event_kind": "code-maintenance-only",
   "timestamp_utc": null,
   "source_url": "https://github.com/Profluent-AI/E1",
   "verification": "verified",
   "description": "The repository changelog records one bug fix in the MSA sampling code and nothing else; no new weights.",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only maintenance commits; no new version or weights",
  "what_it_does": "An encoder that takes either a single sequence or retrieved homologues as context, used for mutation-effect scoring and representation. Its small size and retrieval mode set it apart from the sibling generative model ProGen3; PoET-2 works in the same in-context direction.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "A genomic and metagenomic sequence set curated by the developer, with training reported at about 4 trillion tokens",
    "source_url": "https://doi.org/10.1101/2025.11.12.688125"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": "The paper does not say whether the set contains licensed data",
    "source_url": "https://doi.org/10.1101/2025.11.12.688125"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://github.com/Profluent-AI/E1#comparison-to-other-models",
    "confidence": "medium",
    "second_hand": false,
    "note": "Zero-shot scoring on substitution assays; the repository shows curves rather than per-model values and the official summary table holds no entry"
   },
   {
    "benchmark_id": "bench-15-cameo",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://github.com/Profluent-AI/E1#comparison-to-other-models",
    "confidence": "medium",
    "second_hand": false,
    "note": "Post-hoc use of that continuous evaluation's targets for unsupervised contact maps, curves only"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://www.biorxiv.org/content/10.1101/2025.11.12.688125v1.full",
    "confidence": "medium",
    "second_hand": false,
    "note": "Long-range contact-map evaluation on one CASP round's targets; the values sit in a paper table and were not retrieved this round"
   }
  ],
  "other_evaluations": [
   {
    "label": "Comparison with similar-size protein language models",
    "category": "comparison-vs-baseline-only",
    "detail": "The repository plots size against metric alongside two open encoders and two in-context models",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: the single-sequence mode beats two common open encoders at similar size and the retrieval mode surpasses two in-context models; the public material gives curves, not values.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No entry for the model was found in the official leaderboard file as of 2026-11-16, and no third-party evaluation was found",
  "background": {
   "org_profile": "The developer is an AI-native protein and gene-editing company founded in 2022 in California, reported at about 20 people in 2024 with no current public headcount.",
   "funding": "A $35M round in March 2024 and $106M on 19 November 2025, about $150M in total; the valuation is not officially disclosed, with media reporting it near $1B.",
   "products_partnerships": "The line-up includes a generative protein language model, this encoder and an open gene-editing protein; on 28 April 2026 a multi-programme recombinase collaboration with a large US pharmaceutical company was announced, with milestones up to $2.25B and an undisclosed upfront.",
   "related_models": [
    "model-progen3",
    "model-poet-2",
    "model-esm-c"
   ]
  },
  "reading_notes": [
   "Paper and repository give curves only, so claims of beating another model read as directional, not as checkable numbers.",
   "Retrieval-augmented and single-sequence models are not directly comparable on the same board; the mode must be matched first.",
   "The contact-map work reuses released targets from a continuous evaluation and one CASP round after the fact; it is not participation.",
   "The training set is a privately curated collection, so its composition cannot be reproduced externally."
  ],
  "caveats": [
   "The first-release date comes from the preprint; the DOI request failed this round, so it is marked unverified.",
   "The public material carries no per-model values, so no comparison can be rechecked.",
   "The official leaderboard file holds no entry for the model.",
   "Whether the training set contains licensed data is not disclosed.",
   "The latest event is a code bug fix, not a new version."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2025.11.12.688125",
    "title": "Preprint 2025",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.11.12.688125v1.full",
    "title": "Preprint full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/Profluent-AI/E1",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/Profluent-Bio/E1-600m",
    "title": "Weights repository (600M)",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "release_milestones.code_release",
   "release_milestones.weights_release",
   "open_status.weights_license",
   "training_data.b_private_licensed",
   "benchmarks_reported"
  ],
  "as_of": "2026-11-16",
  "global_entity": "profluent",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-profluent-e1/",
   "en": "/bio/benchmark/models/profluent-e1/"
  },
  "names_full_null_reason": "The paper refers to it as E1 and gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1,
    "retrospective-dataset-use": 2
   }
  },
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-progen3",
  "kind": "model",
  "short_slug": "progen3",
  "names": {
   "display": "ProGen3",
   "full": null,
   "aliases": [
    "ProGen3",
    "progen3-3b"
   ],
   "ambiguous_aliases": [
    "ProGen"
   ]
  },
  "family": "ProGen",
  "family_null_reason": null,
  "domain": "protein",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Sparse mixture-of-experts autoregressive protein language model focused on scaling laws and alignment with laboratory data",
  "org": {
   "names": [
    "Profluent"
   ],
   "global_entity_ids": [
    "profluent"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Protein language model (autoregressive)",
  "open_status": {
   "code": "partial",
   "weights": "non-commercial",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Small and mid sizes are downloadable from the model hub under non-commercial terms while the largest size is unreleased; the site's \"weights not open\" needs correcting, and licence text was not re-checked item by item"
  },
  "first_release": {
   "date": "2025-04-16",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1",
   "event_kind": "preprint-posted",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2025.04.15.649055",
   "verification": "verified",
   "note": "Weights were uploaded to the model hub on 2025-03-20; per this round's rule a repository or hub creation date is not used as first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-04-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.04.15.649055",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-04-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.04.15.649055",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2025-03-20",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights (112M-3B) uploaded to the model hub",
    "event_kind": "weights-release",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/Profluent-Bio/progen3-3b",
    "verification": "partial",
    "note": "The largest size was not released"
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-04-16",
   "precision": "day",
   "year_hint": null,
   "basis": "No new version found after the preprint",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/Profluent-AI/progen3",
   "verification": "verified",
   "description": "The last repository commit, on 2025-07-21, changed a build file and is not a new version; no journal publication or successor model was found (as of 2026-11-16)",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only build-file level changes in the last 12 months, no new version",
  "what_it_does": "Generates protein sequences autoregressively with a sparse mixture-of-experts architecture, studies parameter and data scaling systematically, and then aligns on limited laboratory data, aiming at expressible and functional proteins. It divides work with the same developer's Profluent-E1, which encodes and scores rather than generates, and sits on the same line as generative protein language models such as ESM3.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "An in-house protein atlas v1: 3.4 billion full-length proteins and 1.1 trillion amino-acid tokens from genomes and metagenomes, cleaned in house",
    "source_url": "https://doi.org/10.1101/2025.04.15.649055"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://doi.org/10.1101/2025.04.15.649055"
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "Alignment used \"limited laboratory data\" on activity, expression, stability and binding; its size and provenance are undisclosed",
    "source_url": "https://www.profluent.bio/showcase/progen3"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Separate synthesis, expression and antibody-candidate binding and developability tests were run; counts undisclosed",
    "source_url": "https://www.profluent.bio/showcase/progen3"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Expression and antibody-candidate tests on generated sequences",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "The showcase page gives no design count or denominator",
    "success_rate": null,
    "affinity_best": null,
    "version": "showcase page",
    "source_type": "self",
    "source_url": "https://www.profluent.bio/showcase/progen3",
    "date": {
     "date": "2025-04-16",
     "precision": "day",
     "year_hint": null,
     "basis": "Contemporaneous with the preprint",
     "event_kind": "preprint-posted",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1101/2025.04.15.649055",
     "verification": "partial",
     "note": null
    },
    "verification": "unverified",
    "executor": "Executor undisclosed",
    "unverified_marker": "Scale undisclosed; no figures given",
    "note": "The developer also reports a rise in correlation between predictions and measured fitness after alignment; it is self-reported with no per-item data and is not independent evidence"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "confidence": "high",
    "second_hand": false,
    "note": "The official zero-shot substitution summary table lists several sizes; the largest size is absent"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own scaling study",
    "category": "own-task-set",
    "detail": "Compares generation diversity and expressibility across sizes on in-house data; a developer-defined task set",
    "source_type": "self"
   },
   {
    "label": "Alignment against laboratory data",
    "category": "wet-lab-validation",
    "detail": "Checks alignment against measured fitness; the data and its scale are not public, so it cannot be re-checked externally",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that larger models generate a wider range of expressible, functional proteins and can be aligned with a small amount of laboratory data.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The official zero-shot board lists several of its sizes, an organiser-maintained third-party record.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The developer is an AI-native protein and gene-editing company founded in California in 2022, reported at about 20 staff in 2024 with no current public headcount.",
   "funding": "A $35M round in 2024-03 led by one venture firm and a $106M round on 2025-11-19 co-led by two investors bring the total to about $150M; no valuation is officially disclosed.",
   "products_partnerships": "The product line covers this model, Profluent-E1 and an open gene-editing protein; on 2026-04-28 it signed a multi-programme recombinase collaboration with a large pharmaceutical company, with a disclosed milestone ceiling but an undisclosed upfront.",
   "related_models": [
    "model-profluent-e1",
    "model-esm3"
   ]
  },
  "reading_notes": [
   "Its focus is generation while the board that lists it measures zero-shot scoring; the two are not the same thing.",
   "Adjacent sizes sit close together on that board, so the paper's scaling conclusion cannot be checked there.",
   "The scale of the alignment data is undisclosed, so the reported gain cannot be reproduced independently.",
   "Training used a privately curated large corpus, so overlap with evaluation sets cannot be checked externally."
  ],
  "caveats": [
   "The site stated weights were not open; in fact small and mid sizes are downloadable and only the largest is unreleased, so the statement must be split by size.",
   "The wet-lab work is described only on the showcase page without counts; it is treated as unverified and no figures are given.",
   "The largest size is not on the official board, so any standing claimed for it has no source."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2025.04.15.649055",
    "title": "Preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.profluent.bio/showcase/progen3",
    "title": "Developer showcase page",
    "kind": "official-page",
    "date": null
   },
   {
    "url": "https://huggingface.co/Profluent-Bio/progen3-3b",
    "title": "Model hub weights page",
    "kind": "model-hub",
    "date": null
   },
   {
    "url": "https://github.com/Profluent-AI/progen3",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "Official zero-shot summary table",
    "kind": "leaderboard",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "open_status.weights",
    "was": "Weights not open (licence / early API)",
    "now": "Small and medium weights publicly downloadable on the model hub (non-commercial-type terms); the largest size is not released",
    "source_url": "https://huggingface.co/Profluent-Bio/progen3-3b",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "Wording that there was no later event",
    "now": "Changed to \"no later sourced event found (as of 2026-11-16)\", noting that the 2025-07-21 commit only changed build files",
    "source_url": "https://github.com/Profluent-AI/progen3",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.weights_license",
   "open_status.code_license",
   "wetlab_validation",
   "training_data.c_own_wetlab_for_training.detail"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-progen3/",
   "en": "/bio/benchmark/models/progen3/"
  },
  "names_full_null_reason": "The name is the product name; the paper gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": "No model page in the site-wide entity library; only the developer organisation has one",
  "status_head": null
 },
 {
  "id": "model-prosst",
  "kind": "model",
  "short_slug": "prosst",
  "names": {
   "display": "ProSST",
   "full": null,
   "full_null_reason": "Paper and repository use the short name only",
   "aliases": [
    "ProSST",
    "ProSST-2048"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Different structure-vocabulary sizes (K) are configurations of one model, not members",
  "domain": "protein",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Protein language model quantizing local structure into tokens with disentangled attention",
  "org": {
   "names": [
    "Shanghai Jiao Tong University",
    "Shanghai AI Laboratory"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Protein language model (quantized structure tokens)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code and weights public in the official repository; licence text not retrieved this round, so terms are left empty"
  },
  "first_release": {
   "date": "2024-04-17",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv preprint v1 (the conference version came later)",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2024.04.15.589672",
   "verification": "partial",
   "note": "The preprint DOI is inferred from a title match (unverified); the site previously used the conference publication as first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-04-17",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1 posted",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2024.04.15.589672",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2024-04-17",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2024.04.15.589672",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2024-12",
    "precision": "month",
    "event_kind": "conference-publication",
    "basis": "Formal publication at a 2024 machine-learning conference (exact day not pinned)",
    "timestamp_utc": null,
    "source_url": "https://github.com/ai4protein/ProSST",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2024-12",
   "precision": "month",
   "event_kind": "conference-publication",
   "basis": "Publication at a 2024 machine-learning conference",
   "description": "Formal conference publication; the 2026-07 and 2026-09 README items are project news, not model updates",
   "timestamp_utc": null,
   "source_url": "https://github.com/ai4protein/ProSST",
   "verification": "partial",
   "note": "The exact publication day is not pinned, so it is kept to the month (unverified)"
  },
  "status": "maintained",
  "status_note": "Repository maintained with project news, no new ProSST version",
  "what_it_does": "A quantization module maps each residue's local 3D environment to a discrete structure token, then sequence-structure disentangled attention is pretrained for zero-shot mutation-effect prediction. It is the base model of VenusREM; unlike SaProt the structure vocabulary is learned rather than taken from an off-the-shelf alignment tool.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 18.8M public structures",
    "source_url": "https://github.com/ai4protein/ProSST"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data reported",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "zero-shot substitution DMS summary table, several structure-vocabulary sizes",
    "source_url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "confidence": "high",
    "verification": "verified",
    "note": "The official summary table lists several configurations directly; it leads the sequence-plus-structure category but is not top of the overall board"
   },
   {
    "benchmark_id": "bench-34-abbibench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "antibody binding-affinity correlation, structure-aware LM category",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "This pairing rests only on the site's own page; no primary document was opened (unverified)"
   }
  ],
  "other_evaluations": [
   {
    "label": "Inherited weakness on viral proteins",
    "category": "public-benchmark-not-on-site",
    "detail": "A board maintainer post notes that models inheriting existing embeddings drop on viral-protein subsets, this one included",
    "source_type": "third"
   },
   {
    "label": "Retrieval-augmented successor as a control",
    "category": "comparison-vs-baseline-only",
    "detail": "Used repeatedly as the control for VenusREM, a comparison by the same group",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that learned quantized structure tokens plus disentangled attention put a roughly 110M-parameter model in the top tier for zero-shot mutation-effect prediction.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent re-evaluation or third-party re-run found",
  "background": {
   "org_profile": "An academic team with no separate company: Shanghai Jiao Tong University working with the Shanghai AI Laboratory and further university partners. The work appeared at a 2024 machine-learning conference and grew into a protein-engineering toolchain.",
   "funding": null,
   "funding_null_reason": "Paper acknowledgements not retrieved; funding left blank",
   "products_partnerships": "The same group's Venus line adds retrieval augmentation on top of it to form VenusREM, alongside a protein-engineering system and a mining tool; it shares the structure-aware route with SaProt.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-venusrem",
    "model-saprot",
    "model-poet-2"
   ]
  },
  "reading_notes": [
   "It leads the sequence-plus-structure category on the official board, but time-stamped \"first open model past a threshold\" claims are not carried here.",
   "It depends on predicted structures, and models inheriting existing embeddings generally drop on viral proteins.",
   "No wet lab: it speaks to DMS ranking, not to designability.",
   "The latest update is the conference publication; later repository notes are project news, not a new version."
  ],
  "caveats": [
   "The AbBiBench pairing rests only on this site's page (unverified).",
   "The conference date is month-precision and the preprint DOI is inferred from a title match (unverified).",
   "Retrieval and structure inputs raise scores, so it is not directly comparable with sequence-only models."
  ],
  "sources": [
   {
    "url": "https://github.com/ai4protein/ProSST",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2024.04.15.589672",
    "title": "ProSST preprint (DOI inferred from title match)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "ProteinGym official summary table",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://pascalnotin.substack.com/p/have-we-hit-the-scaling-wall-for",
    "title": "Board maintainer blog post (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.event_kind",
    "was": "Recorded as none-since-release",
    "now": "Changed to conference publication (2024-12)",
    "source_url": "https://github.com/ai4protein/ProSST",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "Said the repository only had README news and no new version, contradicting the recorded date's event",
    "now": "States that 2024-12 is the conference publication and that the 2026 README items are project news, not model updates",
    "source_url": "https://github.com/ai4protein/ProSST",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "Used the conference publication as first release",
    "now": "Changed to the 2024-04-17 preprint v1, the earlier provable public evidence",
    "source_url": "https://doi.org/10.1101/2024.04.15.589672",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "latest_update",
   "benchmarks_reported[1]",
   "release_milestones.official_announcement",
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-prosst/",
   "en": "/bio/benchmark/models/prosst/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-proteina-complexa",
  "kind": "model",
  "short_slug": "proteina-complexa",
  "names": {
   "display": "Proteina-Complexa",
   "full": null,
   "aliases": [
    "Proteina-Complexa"
   ],
   "ambiguous_aliases": [
    "Proteina"
   ]
  },
  "family": null,
  "family_null_reason": "The predecessors Proteina / La-Proteina and the downstream enzyme-design model have their own records; no family entry",
  "domain": "protein",
  "domains_secondary": [
   "generation",
   "small-molecule"
  ],
  "one_liner": "All-atom binder design model combining generative pretraining with test-time compute",
  "org": {
   "names": [
    "NVIDIA (GenAIR)"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (all-atom binder generation + test-time optimisation)",
  "open_status": {
   "code": "open",
   "weights": "research-only",
   "weights_license": "NVIDIA Open Model License (model-catalogue page also carries a proprietary header; wording inconsistent)",
   "code_license": "Apache-2.0 announced on the code page; repository README points to LICENSE only",
   "license_changed": null,
   "note": "Training code and configs are open and checkpoints come from the company catalogue; licence wording differs between pages and the synthetic dataset is released as dimer ID lists only"
  },
  "first_release": {
   "date": "2026-01-26",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer research page publication (conference oral paper)",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://research.nvidia.com/publication/2026-01_proteina-complexa-scaling-atomistic-protein-binder-design-generative",
   "verification": "verified",
   "note": "Earliest sourced public evidence; the method paper's arXiv v1 is 2026-03-30"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-01-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Research page publication",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://research.nvidia.com/publication/2026-01_proteina-complexa-scaling-atomistic-protein-binder-design-generative",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-03-30",
    "precision": "day",
    "year_hint": null,
    "basis": "Method paper arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2603.27950",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-03-16",
    "precision": "day",
    "year_hint": null,
    "basis": "GitHub 1.0.0 commit",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/NVIDIA-BioNeMo/Proteina-Complexa",
    "verification": "partial",
    "note": "The site recorded 03-15; the difference is timezone-level"
   },
   "weights_release": {
    "date": "2026-03-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Checkpoints on the model catalogue (binder / ligand / enzyme-motif sets)",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/models/proteina_complexa",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-01-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Research page",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://research.nvidia.com/publication/2026-01_proteina-complexa-scaling-atomistic-protein-binder-design-generative",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09-18",
   "precision": "day",
   "year_hint": null,
   "basis": "Wet-lab preprint posted (separate from the method paper)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
   "verification": "verified",
   "description": "The wet-lab campaign preprint went up; earlier distinct events are the method paper v1 (2026-03-30), code and checkpoints (2026-03-16) and the research page (2026-01-26). Commits through 2026-11-11 are install and environment changes",
   "note": "Previously recorded as a code release; kind corrected to preprint posted"
  },
  "status": "active",
  "status_note": "Preprint and code activity within 12 months",
  "what_it_does": "Unifies generative modelling with hallucination-style sequence optimisation, emitting binder sequence and structure directly without a separate inverse-folding step and spending extra compute at inference to raise hit rates; staged training (monomers, synthetic dimers, PDB multimers, protein-ligand pairs) with extensions to small-molecule targets and enzyme motif design. Like BoltzGen it is open generative binder design; compared with RFdiffusion2/3 it leans on test-time compute and synthetic data scale.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "The predecessor's 46M structure-sequence pairs, PDB multimers, PLINDER protein-ligand pairs, plus a synthetic dimer dataset",
    "source_url": "https://arxiv.org/abs/2603.27950"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No claim of feeding wet-lab results back into training",
    "source_url": "https://arxiv.org/abs/2603.27950"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Large partner wet-lab campaign used for validation only",
    "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "All-to-all multiplexed phage display",
    "targets": 127,
    "designs_tested": 1015870,
    "hits": 115,
    "denominator_basis": "1,015,870 designed sequences screened against 127 targets in one multiplexed campaign",
    "success_rate": "on-target 115/127 (91%); target-specific 92/127 (72%); poly-specific 78/127 (61%)",
    "affinity_best": null,
    "version": "wet-lab preprint (posted 2026-09-18), not the method paper",
    "source_type": "self",
    "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
    "date": {
     "date": "2026-09-18",
     "precision": "day",
     "year_hint": null,
     "basis": "Wet-lab preprint posted",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Partner-lab executed; not independent replication",
    "unverified_marker": null,
    "note": "Specific and poly-specific sets are not mutually exclusive; 133 is the programme-wide total that contains this 127-target panel and must never be added to it"
   },
   {
    "campaign": "Compute-matched comparison (computational and experimental mix)",
    "targets": 83,
    "designs_tested": 32469,
    "hits": 702,
    "denominator_basis": "83 targets, 32,469 codesigned sequences; 702 on-target hits of which 607 specific",
    "success_rate": "1.87%",
    "affinity_best": null,
    "version": "method paper",
    "source_type": "self",
    "source_url": "https://arxiv.org/abs/2603.27950",
    "date": {
     "date": "2026-03-30",
     "precision": "day",
     "year_hint": null,
     "basis": "Method paper v1",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2603.27950",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Developer-reported",
    "unverified_marker": null,
    "note": "Comparator numbers in the same table are the developers' own runs, not independent re-runs"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-28-adaptyv",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
    "confidence": "high",
    "second_hand": false,
    "note": "The project page states participation and links the organiser page, which does not name the model; recorded as self-reported participation the organiser has not confirmed"
   }
  ],
  "other_evaluations": [
   {
    "label": "Enzyme motif design task set (41 tasks)",
    "category": "public-benchmark-not-on-site",
    "detail": "The paper reports 41 atomic-motif enzyme tasks against an earlier design model; that task set is outside the 44",
    "source_type": "self"
   },
   {
    "label": "Own all-to-all wet-lab panel",
    "category": "wet-lab-validation",
    "detail": "The 127-target panel is a developer-built readout, not a public benchmark",
    "source_type": "self"
   },
   {
    "label": "Partner campaign (programme-wide 133-target framing)",
    "category": "case-study",
    "detail": "The project page cites over a million candidates across 133 targets; the 127-target panel is a subset",
    "source_type": "self"
   },
   {
    "label": "Compute-matched in-silico success comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "The abstract claims higher in-silico success than prior generative and hallucination methods at matched compute",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: pretraining on synthetic dimer data plus test-time compute gives higher in-silico success than prior methods, with on-target hits for 91% of 127 targets in a pooled screen.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Wet-lab work was partner-executed and developer-compiled; no independent evaluation found",
  "background": {
   "org_profile": "A generative-AI research group inside a listed GPU and AI platform company (Santa Clara, California); its life-science model line sits under the company's open digital-biology platform and is funded from corporate R&D with no separate financing. External co-author institutions include two research institutes in the UK and Canada and a Korean national university.",
   "funding": null,
   "products_partnerships": "Wet-lab partners include a multiplexed protein measurement company, a CRO, a large pharma company and four university labs, mostly validation collaborations with undisclosed terms; a downstream enzyme-design model with an academic institute adds explicit backbone coordinates and cuts sampling steps from about 400 to 60.",
   "related_models": [
    "model-la-proteina",
    "model-rfdiffusion2-3",
    "model-boltzgen"
   ]
  },
  "reading_notes": [
   "127 and 133 are different things: 127 is the pooled panel, 133 the programme-wide total that contains it; cite hits as 115/127.",
   "Wet-lab figures must cite the wet-lab preprint, not the method paper; they are separate events.",
   "A hit means at least one binding design per target, not a per-design rate, and the result depends on test-time compute, so it is not directly comparable with single-shot models.",
   "Adaptyv participation is self-reported and unconfirmed by the organiser; MotifBench is not mentioned in the paper and PLINDER appears only as training data."
  ],
  "caveats": [
   "The only site pairing is self-reported participation the organiser has not confirmed.",
   "The large wet-lab campaign is validation, not training; it is not a data flywheel.",
   "Licence wording differs between the code page and the model catalogue page.",
   "The benefit of the synthetic-data route rests on developer reports."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2603.27950",
    "title": "Method paper, arXiv 2603.27950",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
    "title": "Project page (with wet-lab preprint)",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://research.nvidia.com/publication/2026-01_proteina-complexa-scaling-atomistic-protein-binder-design-generative",
    "title": "Research publication page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/NVIDIA-BioNeMo/Proteina-Complexa",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/models/proteina_complexa",
    "title": "Model catalogue checkpoint page",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "92 target-specific (no denominator), multispecific omitted",
    "now": "115/127, 92/127, 78/127, citing the wet-lab preprint",
    "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2026-03-30, code release",
    "now": "2026-09-18, wet-lab preprint online",
    "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Zero pairings",
    "now": "Adaptyv modified-protocol self-report; the model leaves the zero-pairing list",
    "source_url": "https://research.nvidia.com/labs/genair/proteina-complexa/",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Candidate pairings MotifBench, PLINDER",
    "now": "Neither is a pairing: the paper does not mention the first, and the second is used only as training data",
    "source_url": "https://arxiv.org/abs/2603.27950",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.code_license",
   "open_status.weights_license",
   "release_milestones.code_release",
   "release_milestones.weights_release"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-proteina-complexa/",
   "en": "/bio/benchmark/models/proteina-complexa/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-protenix",
  "kind": "model",
  "short_slug": "protenix",
  "names": {
   "display": "Protenix",
   "full": null,
   "full_null_reason": "Official materials use Protenix; the companion tools are named PXDesign and PXMeter",
   "aliases": [
    "Protenix",
    "Protenix-v1",
    "Protenix-v2"
   ],
   "ambiguous_aliases": [
    "Protenix-Mini"
   ]
  },
  "family": null,
  "family_null_reason": "The v0.5 to v2 line is one record; PXDesign and PXMeter are companion tools and SeedFold has its own record",
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "Open co-folding series reproducing and trying to exceed the contemporary closed model, with v2 weights turned proprietary",
  "org": {
   "names": [
    "ByteDance Seed",
    "ByteDance Seed AI for Science"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Co-folding (plus the design tool PXDesign)",
  "open_status": {
   "code": "partial",
   "weights": "closed",
   "weights_license": "v1 weights open; v2 weights proprietary, not under an open licence",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code is Apache-2.0; v1 weights are open while v2 weights are proprietary and not released under an open licence, so third parties cannot re-check the new version"
  },
  "first_release": {
   "date": "2024-12-12",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "First GitHub release v0.1.0; preprint 2025-01-11; the repository creation date is excluded by policy",
   "timestamp_utc": null,
   "source_url": "https://github.com/bytedance/Protenix/releases",
   "verification": "partial",
   "note": "A possible 2024-11 announcement was not checked this round (partially verified)"
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-01-11",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.01.08.631967",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-12-12",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "First GitHub release v0.1.0 (v0.3.2 on 2024-12-18)",
    "timestamp_utc": null,
    "source_url": "https://github.com/bytedance/Protenix/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2024-12-12",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "The v1-line weights were open with the code",
    "timestamp_utc": null,
    "source_url": "https://github.com/bytedance/Protenix",
    "verification": "partial",
    "note": "The v2 weights are proprietary and outside this milestone"
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-04-07",
   "precision": "day",
   "event_kind": "new-version",
   "basis": "GitHub release v2.0.0",
   "description": "v2 released with improved antibody-antigen prediction and proprietary weights not under an open licence; later commits are docs and maintenance",
   "timestamp_utc": null,
   "source_url": "https://github.com/bytedance/Protenix/releases",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "A new version in 2026-04 with maintenance commits since",
  "what_it_does": "An open reproduction and extension of AlphaFold 3 aiming to match or beat it under the same data cutoff, with about 368M parameters in v1 and 464M in v2 plus template and RNA alignment features. Unlike OpenFold3 it uses public data only, and unlike Boltz-2 it has no affinity head.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB (cutoff 2021-09-30, plus a 2025-06-30 version), alignments and distillation data; the training data and alignment pipeline are open",
    "source_url": "https://github.com/bytedance/Protenix"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Wet lab validates only the companion design tool PXDesign; the structure-prediction side has none",
    "source_url": "https://protenix.github.io/pxdesign/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "PXDesign binder campaign",
    "modality": "protein binder",
    "targets": 7,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Seven targets, six of which yielded nanomolar hits; per-target denominators were not opened in primary material",
    "success_rate": "nanomolar hits on 6/7 targets; hit rates 17-82%",
    "affinity_best": null,
    "version": "design preprint v3 (2025-12-17)",
    "source_type": "self",
    "source_url": "https://protenix.github.io/pxdesign/",
    "date": {
     "date": "2025-12-17",
     "precision": "day",
     "event_kind": "preprint-posted",
     "basis": "Design preprint v3",
     "timestamp_utc": null,
     "source_url": "https://protenix.github.io/pxdesign/",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "The range published earlier came from a stale repository README; corrected to the design preprint and cited to the design paper rather than the model repository; developer-reported, not an independent replication"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "official set, early version only (via the Mini paper)",
    "source_url": "https://arxiv.org/html/2507.11839v1",
    "confidence": "low",
    "verification": "partial",
    "note": "Holds only for an early version: neither the v1 nor the v2 report scores on that test set, v1 uses only its validity criteria as metrics, and the site matrix overstates it for current versions"
   },
   {
    "benchmark_id": "bench-12-dude",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": null,
    "source_url": null,
    "confidence": "low",
    "verification": "unverified",
    "note": "The basis is the site's own page (unverified): a second-hand mention only, with no values"
   },
   {
    "benchmark_id": "bench-30-plinder",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "v1 self-report on the subset where all compared models have results (figure only)",
    "source_url": "https://arxiv.org/abs/2510.24670",
    "confidence": "medium",
    "verification": "partial",
    "note": "v1 gives only a figure binned by similarity to training with no summary value; a third-party paper supplies a summary figure"
   },
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer leaderboard",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "high",
    "verification": "verified",
    "note": null
   },
   {
    "benchmark_id": "bench-33-posex",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "self-docking set, benchmark authors' run",
    "source_url": "https://arxiv.org/html/2505.01700v3",
    "confidence": "medium",
    "verification": "partial",
    "note": "Re-run by the benchmark's authors"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own ligand evaluation set (companion tool)",
    "category": "own-task-set",
    "detail": "A 623-entry in-house set scored with public validity criteria after manual removal of experimental artefacts; scores are not shown here",
    "source_type": "self"
   },
   {
    "label": "The design-side tool PXDesign",
    "category": "wet-lab-validation",
    "detail": "The wet-lab validation belongs to the companion design tool and is not a result of the structure-prediction model",
    "source_type": "self"
   },
   {
    "label": "Third-party tables and second-hand paraphrase",
    "category": "third-party-evaluation",
    "detail": "Several third-party tables list v1 alongside the contemporary closed model, partly relayed by second-hand platforms under inconsistent protocols",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: parity with the contemporary closed model under the same data cutoff, clearly improved antibody-antigen prediction in v2, and five seeds achieving what v1 needed a thousand seeds for.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Third-party re-runs (the benchmark's authors and one comparison paper) place it below the open controls it is compared with, and with v2 weights proprietary later versions cannot be re-checked by third parties.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "It comes from a large internet company's foundation-model research division, whose science agenda spans biological foundation models, quantum chemistry and molecular dynamics, funded from the company's internal budget.",
   "funding": null,
   "funding_null_reason": "Big-tech internal budget, no separate fundraising",
   "products_partnerships": "The series and its companion design and evaluation tools are open, and the same company also has SeedFold; a 2026-06 report says it is spinning out an AI drug-discovery unit of about 50 people with the structure team joining, without stating whether this series moves (second-hand, unverified).",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-alphafold3",
    "model-openfold3",
    "model-rf3",
    "model-boltz-2",
    "model-seedfold"
   ]
  },
  "reading_notes": [
   "On the maintainer board v1 sits in the same band as the contemporary closed model under official framing, which supports parity but not superiority.",
   "With v2 weights closed, later versions rest on official figures alone and cannot be re-checked by third parties.",
   "The only summary figure available on the site comes from a third-party paper, as the official v1 report gives binned figures only.",
   "The developers prefer their own off-site evaluations, which sit closer to antibody and RNA use cases and must not be mixed with site entries."
  ],
  "caveats": [
   "The v2 weights are proprietary, so openness must be read separately for code and weights.",
   "The public pose-set pairing holds only for an early version and must carry a version qualifier.",
   "The wet-lab figures belong to the companion design tool, not to the structure-prediction model.",
   "One pairing rests only on the site's own page (unverified).",
   "The report on the company's spin-out is second-hand and does not say whether Protenix moves with it (unverified)."
  ],
  "sources": [
   {
    "url": "https://github.com/bytedance/Protenix",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/bytedance/Protenix/releases",
    "title": "Official repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.01.08.631967",
    "title": "The model's preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://protenix.github.io/pxdesign/",
    "title": "Design tool preprint and project page",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2507.11839v1",
    "title": "The team's own compact-version paper",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2510.24670",
    "title": "Third-party re-run paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2505.01700v3",
    "title": "Re-run paper by the benchmark's authors",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://kr-asia.com/bytedance-spins-out-drug-discovery-unit-to-test-ai4s-commercialization",
    "title": "Coverage of the company's spin-out (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "First release taken as the preprint date 2025-01-11",
    "now": "Changed to the first code release on 2024-12-12; the repository creation date is excluded by policy",
    "source_url": "https://github.com/bytedance/Protenix/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "The v2 release was dated 2026-04-08",
    "now": "Changed to 2026-04-07 per the release page; later commits are documentation",
    "source_url": "https://github.com/bytedance/Protenix/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "The design-side wet-lab hit-rate range was 20-73%, cited to the model repository",
    "now": "Changed to the design preprint v3 range of 17-82% with nanomolar hits on 6 of 7 targets, cited to the design paper rather than the model repository",
    "source_url": "https://protenix.github.io/pxdesign/",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "The public pose-set cell was shown as a self-report for the current version",
    "now": "Version-qualified: it holds only for an early version, as neither v1 nor v2 reports on that test set",
    "source_url": "https://arxiv.org/html/2507.11839v1",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "release_milestones.weights_release",
   "benchmarks_reported[1]",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-protenix/",
   "en": "/bio/benchmark/models/protenix/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-quine",
  "kind": "model",
  "short_slug": "quine",
  "names": {
   "display": "Quine",
   "full": null,
   "aliases": [
    "Microsoft Quine"
   ],
   "ambiguous_aliases": [
    "Quine"
   ]
  },
  "family": null,
  "family_null_reason": "A single system with no later version",
  "domain": "text-agents",
  "domains_secondary": [
   "single-cell"
  ],
  "one_liner": "Microsoft Research's biology world model plus research harness, research use only",
  "org": {
   "names": [
    "Microsoft Research"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Agent / world model + harness",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Initially limited to the Quine Fellows programme and selected collaborations, research use only"
  },
  "first_release": {
   "date": "2026-09-29",
   "precision": "day",
   "year_hint": null,
   "basis": "Microsoft Research official blog",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-09-29",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-09-29",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09-29",
   "precision": "day",
   "year_hint": null,
   "basis": "No update found after launch",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
   "verification": "verified",
   "description": "No later sourced event found (as of 2026-11-16); access is limited to Quine Fellows and selected collaborations, with broader access via Microsoft Discovery planned without a timeline",
   "note": null
  },
  "status": "product-only",
  "status_note": "Closed, restricted access",
  "what_it_does": "Two parts: a multimodal world model spanning genomics, proteins, chemistry, RNA, cell state and bioimaging, and a harness linking models, scientific tools, literature and researchers; Quine makes predictions that are tested in the wet lab. Compared with hypothesis-focused Co-Scientist and tool-orchestrating Biomni, it ties its own predictive models to a wet-lab partner within one research programme; for cell-state prediction compare STATE.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "No datasets named",
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/"
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "Pancreatic-cancer patient-derived ex vivo models come from a multi-year collaboration with Broad; whether they were used for training is not stated",
    "source_url": "https://www.microsoft.com/en-us/research/project/project-ex-vivo/"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Pancreatic-cancer cell-state predictions were tested in ex vivo experiments on the Broad side",
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "PDAC cell-state transition",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "No counts disclosed: the blog says only that \"thousands\" of compounds were ranked and \"a handful\" tested",
    "success_rate": null,
    "affinity_best": null,
    "version": "developer blog",
    "source_type": "self",
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
    "date": {
     "date": "2026-09-29",
     "precision": "day",
     "year_hint": null,
     "basis": "Official blog",
     "event_kind": "product-update",
     "timestamp_utc": null,
     "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
     "verification": "verified",
     "note": null
    },
    "verification": "unverified",
    "executor": "Ex vivo experiments at partner Broad Institute; not independent replication",
    "unverified_marker": "unverified",
    "note": "The blog says top-ranked compounds gave the largest intended state shifts and others were predicted and confirmed to push cells to a third phenotype; no hit rate, negatives or controls"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Site text and agent benchmarks",
    "category": "none-disclosed",
    "detail": "Checked LAB-Bench, BixBench, BioASQ, PubMedQA, BioML-bench, BioMysteryBench: the official blog gives no benchmark score",
    "source_type": "self"
   },
   {
    "label": "PDAC cell-state transition prediction",
    "category": "wet-lab-validation",
    "detail": "With Broad, compound rankings tested on patient-derived ex vivo models; counts not disclosed",
    "source_type": "partner"
   }
  ],
  "claimed_advantage": "Developers report: a multimodal biology world model plus research harness narrows candidates quickly (about a weekend), and its top predictions were borne out in wet-lab assays.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Only press commentary; no third-party reproduction or quantitative independent evaluation found",
  "background": {
   "org_profile": "Microsoft Research is Microsoft's research division, with areas such as AI for Science and Health Futures and labs in Redmond, Cambridge (UK and US), New York, Montreal, Bangalore, Beijing, Nairobi and elsewhere.",
   "funding": "Microsoft's internal research budget; no separate funding.",
   "products_partnerships": "Project Ex Vivo is a joint Microsoft and Broad Institute cancer research collaboration supported by Dana-Farber Cancer Institute, initially focused on pancreatic cancer; access runs through Quine Fellows and later Microsoft Discovery.",
   "related_models": [
    "model-co-scientist",
    "model-biomni",
    "model-state"
   ]
  },
  "reading_notes": [
   "There are no numbers, so its margin over other systems cannot be judged.",
   "The PDAC case shows prediction-to-wet-lab confirmation is feasible, but with no hit rate, negatives or controls it says nothing about other diseases.",
   "The weekend covers only narrowing thousands of compounds to a few; the experimental system rests on years of collaboration.",
   "Restricted access makes independent evaluation nearly impossible."
  ],
  "caveats": [
   "The wet-lab work has no counts, is treated as unverified and cannot be turned into a success rate.",
   "Training data are entirely undisclosed.",
   "Positioned as experimental research technology; the developers warn outputs may be incomplete or inaccurate."
  ],
  "sources": [
   {
    "url": "https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/",
    "title": "Microsoft Research official blog",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://www.microsoft.com/en-us/research/project/project-ex-vivo/",
    "title": "Project Ex Vivo page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.microsoft.com/en-us/research/about-microsoft-research/",
    "title": "About Microsoft Research",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "wetlab_validation[0]",
   "training_data.c_own_wetlab_for_training"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-quine/",
   "en": "/bio/benchmark/models/quine/"
  },
  "names_full_null_reason": "A product name with no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the partner institute but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-reasyn-v2",
  "kind": "model",
  "short_slug": "reasyn-v2",
  "names": {
   "display": "ReaSyn v2",
   "full": null,
   "aliases": [
    "ReaSyn v2",
    "NV-ReaSyn-AR-166M-v2",
    "NV-ReaSyn-EB-174M-v2"
   ],
   "ambiguous_aliases": [
    "ReaSyn"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "ReaSyn v1",
     "id_or_null": null,
     "first_release": {
      "date": "2025-09-19",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint v1",
      "event_kind": "paper-revision",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2509.16084",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "ReaSyn v2 (AR-166M)",
     "id_or_null": null,
     "first_release": {
      "date": "2026-01-09",
      "precision": "day",
      "year_hint": null,
      "basis": "First commit of the checkpoint files",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://huggingface.co/nvidia/NV-ReaSyn-EB-174M-v2",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "ReaSyn v2 (EB-174M)",
     "id_or_null": null,
     "first_release": {
      "date": "2026-01-09",
      "precision": "day",
      "year_hint": null,
      "basis": "First commit of the checkpoint files",
      "event_kind": "new-weights",
      "timestamp_utc": null,
      "source_url": "https://huggingface.co/nvidia/NV-ReaSyn-EB-174M-v2",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "This record centres on the second version; the two papers carry different titles and v2 no longer describes the reinforcement-learning step of v1"
  },
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Synthesisability projection model that builds a synthesis route from purchasable building blocks",
  "org": {
   "names": [
    "NVIDIA",
    "KAIST"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "US",
   "KR"
  ],
  "model_class": "Molecule generation (synthesis route prediction)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "NVIDIA Open Model License",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code under Apache-2.0 with two second-version checkpoints in a public model hub under the developer's own open model licence"
  },
  "first_release": {
   "date": "2025-09-19",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2509.16084",
   "verification": "verified",
   "note": "The v2 checkpoints went up in January 2026"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-09-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2509.16084",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-09-19",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2509.16084",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-01-11",
    "precision": "day",
    "year_hint": null,
    "basis": "README update on the default branch",
    "event_kind": "documentation-update",
    "timestamp_utc": null,
    "source_url": "https://github.com/NVIDIA-Digital-Bio/ReaSyn",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-01-09",
    "precision": "day",
    "year_hint": null,
    "basis": "Second-version checkpoint files committed",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/nvidia/NV-ReaSyn-EB-174M-v2",
    "verification": "partial",
    "note": "The model card also gives 2026-01-08 as the checkpoint release date"
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-01-12",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer platform announcement (not reverified this round)",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://github.com/NVIDIA-Digital-Bio/ReaSyn",
    "verification": "unverified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-01-11",
   "precision": "day",
   "year_hint": null,
   "basis": "README edit on the default branch",
   "event_kind": "documentation-update",
   "timestamp_utc": null,
   "source_url": "https://github.com/NVIDIA-Digital-Bio/ReaSyn",
   "verification": "verified",
   "description": "The last substantive change on the default branch is a README update; decoding and index speed-ups exist only on three unmerged pull-request branches, and an October 2026 commit only updates citation details.",
   "note": "The site previously described unmerged branch work as a delivered update"
  },
  "status": "maintained",
  "status_note": "Only documentation and citation updates on the main branch; no new weights",
  "what_it_does": "Generates synthesis routes as reasoning chains with an encoder-decoder, finding, for an unsynthesisable molecule, a close analogue assembled from catalogue building blocks and reaction templates; the second version adds bottom-up and top-down tree search plus a discrete-flow model that edits a route by insertion, deletion and substitution. It sits on the same molecule-generation line as GenMol from the same organisation but targets synthesisability rather than general generation.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "211,220 purchasable building blocks from one supplier catalogue and 115 reaction templates, following an existing published setup",
    "source_url": "https://arxiv.org/html/2509.16084v2"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "The 10.5 million in-house training routes are all computationally generated",
    "source_url": "https://arxiv.org/html/2509.16084v2"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-19-pmo",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2509.16084v2",
    "confidence": "medium",
    "second_hand": false,
    "note": "Reported on 15 of that protocol's tasks rather than all of them, against an unconstrained evolutionary baseline and three synthesis-constrained methods"
   }
  ],
  "other_evaluations": [
   {
    "label": "Synthesis route reconstruction",
    "category": "own-task-set",
    "detail": "Route reconstruction on 1,000 molecules from each of three sources, against two synthesis-constrained methods",
    "source_type": "self"
   },
   {
    "label": "Hit expansion and surrogate optimisation",
    "category": "own-task-set",
    "detail": "Analogue expansion and optimisation scored by two computational surrogates, with route accessibility checked by a retrosynthesis tool",
    "source_type": "self"
   },
   {
    "label": "Absent from a third-party comparison",
    "category": "public-benchmark-not-on-site",
    "detail": "A conference paper comparing several models in this area does not list it",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: keeping optimisation on a par with unconstrained methods while enforcing synthesisability, and beating existing synthesis-constrained methods at route reconstruction and hit expansion.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation was found, and a comparison in the same area does not list it",
  "background": {
   "org_profile": "Produced by a US compute company's research team with a South Korean university, with the lead work done during an internship there; the company's life-science business runs through an open development platform and an open model family.",
   "funding": "Internal corporate research with no separate funding.",
   "products_partnerships": "Released alongside an RNA model and several chemistry tool libraries; in January 2026 the company and a US pharmaceutical company announced a joint AI lab with up to $1B over five years, with the split undisclosed.",
   "related_models": [
    "model-genmol",
    "model-saturn",
    "model-reinvent-4"
   ]
  },
  "reading_notes": [
   "Matching an unconstrained optimiser means the constraint costs nothing, not that it beats other optimisers.",
   "Reconstruction depends heavily on where the molecules come from and looks far better on the set that shares a source with the training blocks.",
   "Synthesisable here means definable by templates and a catalogue, not demonstrably makeable; the paper contains no experimental synthesis.",
   "The latest substantive event is a documentation update, with the speed-up code still sitting on unmerged branches."
  ],
  "caveats": [
   "Its pairing here covers only part of that protocol's task set.",
   "Synthesisability is judged by computational tools with no experimental check.",
   "The training routes are entirely computationally generated.",
   "Model card and repository give checkpoint release dates one day apart.",
   "The platform announcement date was not reverified this round."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2509.16084",
    "title": "Preprint v1 2025",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2509.16084v2",
    "title": "Preprint v2 full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/NVIDIA-Digital-Bio/ReaSyn",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/nvidia/NV-ReaSyn-EB-174M-v2",
    "title": "Second-version checkpoint model card",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-07-18 (speed-up commit on an unmerged branch)",
    "now": "2026-01-11 (README update on the default branch); event kind changed to documentation update",
    "source_url": "https://github.com/NVIDIA-Digital-Bio/ReaSyn",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.weights_release",
   "release_milestones.official_announcement",
   "training_data.b_private_licensed"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-reasyn-v2/",
   "en": "/bio/benchmark/models/reasyn-v2/"
  },
  "names_full_null_reason": "A code name plus version, with no expanded form in the paper",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-reinvent-4",
  "kind": "model",
  "short_slug": "reinvent-4",
  "names": {
   "display": "REINVENT 4",
   "full": null,
   "aliases": [
    "REINVENT4",
    "REINVENT 4"
   ],
   "ambiguous_aliases": [
    "REINVENT"
   ]
  },
  "family": "REINVENT",
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Open-source small-molecule generative design framework driven by reinforcement, transfer and curriculum learning",
  "org": {
   "names": [
    "AstraZeneca Molecular AI (Discovery Sciences, R&D)"
   ],
   "global_entity_ids": []
  },
  "org_type": "pharma",
  "org_type_note": null,
  "country": [
   "SE",
   "GB"
  ],
  "model_class": "Molecule generation and optimisation framework",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code and priors are Apache-2.0, with priors hosted in a public data repository from v4.6.22"
  },
  "first_release": {
   "date": "2023-11-07",
   "precision": "day",
   "year_hint": null,
   "basis": "First public repository release (v4.0.11)",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/MolecularAI/REINVENT4/releases",
   "verification": "verified",
   "note": "The site recorded 2024-02-21, which is the journal paper and v4.1.8 date, not the first public version"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-11-07",
    "precision": "day",
    "year_hint": null,
    "basis": "First public release",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/MolecularAI/REINVENT4/releases",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": {
    "date": "2024-02-21",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal paper online",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1186/s13321-024-00812-5",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2023-11-07",
    "precision": "day",
    "year_hint": null,
    "basis": "v4.0.11",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/MolecularAI/REINVENT4/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-07-15",
    "precision": "day",
    "year_hint": null,
    "basis": "Priors moved to a public data repository (v4.6.22)",
    "event_kind": "weights-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/MolecularAI/REINVENT4/releases",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-11-06",
   "precision": "day",
   "year_hint": null,
   "basis": "v4.9 release",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://github.com/MolecularAI/REINVENT4/releases",
   "verification": "verified",
   "description": "v4.9 adds modular diversity filters, a clustering component, intrinsic rewards and a similarity scoring component, and removes deprecated settings",
   "note": null
  },
  "status": "active",
  "status_note": "A new version was released within 12 months",
  "what_it_does": "Built on recurrent and transformer priors with reinforcement learning, driven from the command line with file-based task configuration, it covers de novo design, R-group replacement, linker design, scaffold hopping and optimisation, merging four previously separate tools into one repository. Unlike prompt-style general generators such as GenMol it needs a scoring function defined per task, and it sits opposite Saturn on the same reinforcement-learning line.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Priors were trained unsupervised on SMILES from public molecule libraries; the paper lists no library versions and the versions given by secondary compilations could not be confirmed in the original",
    "source_url": "https://doi.org/10.1186/s13321-024-00812-5"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The paper discloses no internal data used to train the public priors; internal project use is application, not training data",
    "source_url": "https://doi.org/10.1186/s13321-024-00812-5"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1186/s13321-024-00812-5"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1186/s13321-024-00812-5"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-19-pmo",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "evidence_url": "https://arxiv.org/abs/2206.12411",
    "confidence": "medium",
    "second_hand": false,
    "note": "The benchmark paper implements the previous REINVENT generation as a baseline; this version reports no result there, so it is not counted"
   }
  ],
  "other_evaluations": [
   {
    "label": "Feature comparison with two similar tools",
    "category": "comparison-vs-baseline-only",
    "detail": "The paper gives a table of design modes and learning styles, not performance scores",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report one open framework covering several molecule design modes, already used in production on internal projects.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Public evaluations feature the previous generation's implementation; no independent evaluation of this version was found",
  "background": {
   "org_profile": "The developer is a molecular-AI group inside a large pharmaceutical company's R&D organisation, based in Sweden with members in the UK and funded from internal R&D budget.",
   "funding": null,
   "products_partnerships": "The same group also maintains a retrosynthesis planning tool; this framework merges four previously separate tools and its code is in a public repository.",
   "related_models": [
    "model-saturn",
    "model-genmol"
   ]
  },
  "reading_notes": [
   "Its value lies in engineering maturity and production use, not in benchmark standing.",
   "REINVENT figures in public evaluations usually come from a third party's implementation of the previous generation and must not be read as this version's results.",
   "The paper compares with only two similar tools and only by feature table, a narrow scope.",
   "No public wet-lab data corresponds directly to this model."
  ],
  "caveats": [
   "The site's recorded first-release date was the journal paper date and is corrected to the first public version.",
   "Under this round's relation rules it enters the zero-pairing list: it has no countable pairing with any benchmark on this site.",
   "The versions of the public libraries behind the priors have no primary source, so second-hand version numbers should not be repeated."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1186/s13321-024-00812-5",
    "title": "Journal paper",
    "kind": "journal",
    "date": "2024-02-21"
   },
   {
    "url": "https://github.com/MolecularAI/REINVENT4/releases",
    "title": "Release list",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/MolecularAI/REINVENT4",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2206.12411",
    "title": "Third-party benchmark paper",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2024-02-21",
    "now": "2023-11-07 (first public release v4.0.11)",
    "source_url": "https://github.com/MolecularAI/REINVENT4/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Has results on this optimization benchmark (third-party)",
    "now": "Only the previous generation is used by third parties as a baseline; this version has no self-reported result, so the relation is changed to cited only and not counted",
    "source_url": "https://arxiv.org/abs/2206.12411",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "training_data.a_public.detail"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-reinvent-4/",
   "en": "/bio/benchmark/models/reinvent-4/"
  },
  "names_full_null_reason": "The name is a software name; the paper gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "cited-as-comparison-only": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-retrodfm-r",
  "kind": "model",
  "short_slug": "retrodfm-r",
  "names": {
   "display": "RetroDFM-R",
   "full": null,
   "aliases": [
    "RetroDFM-R",
    "RetroDFM-R-8B"
   ],
   "ambiguous_aliases": [
    "ChemDFM"
   ]
  },
  "family": null,
  "family_null_reason": "The sibling chemistry LLM line has no site entity record, so no family entry",
  "domain": "small-molecule",
  "domains_secondary": [
   "text-agents"
  ],
  "one_liner": "Reasoning-style retrosynthesis LLM trained with verifiable chemistry rewards, emitting precursors with a reasoning trace",
  "org": {
   "names": [
    "Shanghai Jiao Tong University (X-LANCE Lab; School of Chemistry and Chemical Engineering)",
    "Suzhou Laboratory"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Retrosynthesis (chemistry LLM + reinforcement-learning reasoning)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "GPL-3.0 (per the group's model pages)",
   "code_license": "GPL-3.0",
   "license_changed": null,
   "note": "Weights and code are GPL-3.0 per the group's public pages; the hosting namespace is the lab's open-source space, not a company"
  },
  "first_release": {
   "date": "2025-07-23",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2507.17448 v1 submitted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2507.17448",
   "verification": "verified",
   "note": "The site previously cited another paper's identifier (2510.16588); corrected"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-07-23",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2507.17448",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-07-23",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1 (v2 on 2026-08-24)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2507.17448",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-08-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Code repository commits including the licence; no release object exists",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/OpenDFM/RetroDFM-R",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2025-11-20",
    "precision": "day",
    "year_hint": null,
    "basis": "First 8B weights repository created (README announcement 2025-11-22)",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/OpenDFM/RetroDFM-R-v0-8B",
    "verification": "partial",
    "note": "Relies on a host-side creation date; the site's 2026-01-12 is only a README edit, and a second 8B repository was created 2026-08-20"
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-08-25",
   "precision": "day",
   "year_hint": null,
   "basis": "Code repository commits (no release object)",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://github.com/OpenDFM/RetroDFM-R",
   "verification": "partial",
   "description": "Code repository commits, alongside the paper's v2 (2026-08-24) and the creation of a new 8B weights repository (2026-08-20). No journal version was found and whether the newer weights change results is unverified. This is the latest sourced event",
   "note": "Previously 2026-08-24; corrected to 2026-08-25 per the repository commits"
  },
  "status": "active",
  "status_note": "Paper revision, weights and code events within 12 months",
  "what_it_does": "Starts from the group's chemistry LLM, continues pretraining on retrosynthesis text, cold-starts with distilled chain-of-thought and then applies reinforcement learning with verifiable signals such as reaction plausibility and product match, emitting precursors together with a reasoning trace. Unlike the pretraining route of Uni-Mol3 or the sequence-to-sequence route of Chemformer, the reasoning text is part of the output, which is its interpretability claim.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Continued pretraining on retrosynthesis chemistry data, including structure-name pairs from a public compound library and public retrosynthesis prediction data",
    "source_url": "https://arxiv.org/html/2507.17448v1"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The paper reports no wet-lab validation",
    "source_url": "https://arxiv.org/html/2507.17448v1"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-07-uspto",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2507.17448v1",
    "confidence": "high",
    "second_hand": false,
    "note": "Two official splits (the small and full sets) in the reaction-class-unknown setting; the settings with and without test-time augmentation must be cited separately, and the baselines in the same table are the authors' own runs"
   }
  ],
  "other_evaluations": [
   {
    "label": "Double-blind preference study",
    "category": "own-task-set",
    "detail": "A double-blind comparison of the model's precursors against ground-truth reactants, with most responses rating the model output as better or equal; disagreements cluster on halogen substituent swaps, and the study was run by the authors",
    "source_type": "self"
   },
   {
    "label": "Multi-step route planning case studies",
    "category": "case-study",
    "detail": "Several multi-step synthesis route case studies with no common metric",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: reinforcement learning with verifiable chemistry rewards beats existing sequence and edit-based methods on the official retrosynthesis splits while producing a readable reasoning trace.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party re-run or independent evaluation was found",
  "background": {
   "org_profile": "An academic lab: a cross-media language intelligence lab in a Chinese university's computer science school, working with the same university's chemistry school and a regional laboratory; its open-source namespace is not a company. The lab also runs a joint lab with a speech-technology company.",
   "funding": "Acknowledgements list a national key programme, three national natural science foundation grants and two municipal science projects; training used eight datacentre-class GPUs.",
   "products_partnerships": "Sibling work includes a chemistry LLM with multimodal and reasoning variants (three public papers); on-site competitors are Chemformer and Uni-Mol3. No company partnerships disclosed.",
   "related_models": [
    "model-chemformer",
    "model-uni-mol3"
   ]
  },
  "reading_notes": [
   "The site cited the wrong paper identifier; the correct source is arXiv 2507.17448 and the other identifier belongs to a different paper.",
   "The site's unknown-class figure is unverified and the abstract gives a different value, so always go back to the abstract's framing.",
   "With and without test-time augmentation are two settings with a visible gap; the figures once recorded as an abstract-versus-text conflict were a single reaction class, not the overall framing.",
   "USPTO data are frozen and leak templates, so high scores are not synthetic feasibility; there is no wet-lab or route-level independent validation."
  ],
  "caveats": [
   "All benchmark results are author-reported with no board entry.",
   "Weight dates rely on host-side repository creation; whether the two 8B weight releases differ in results is unverified.",
   "The reasoning trace aids readability, but its correctness rests on self-assessment only.",
   "The code repository has no release objects, so versioning can only be read from commits."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2507.17448",
    "title": "Model paper, arXiv 2507.17448 (correct source)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2507.17448v1",
    "title": "Paper v1 full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/OpenDFM/RetroDFM-R",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/OpenDFM/RetroDFM-R-v0-8B",
    "title": "First 8B weights repository",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://huggingface.co/OpenDFM/RetroDFM-R-8B",
    "title": "Newer 8B weights repository",
    "kind": "package-registry",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "sources",
    "was": "Cited arXiv 2510.16588 as this model's paper",
    "now": "The correct id is arXiv 2507.17448; the earlier id belongs to another paper",
    "source_url": "https://arxiv.org/html/2507.17448v1",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "The \"unknown class\" retrosynthesis figure shown on the site",
    "now": "The figure is not confirmed; recorded per the abstract, and this page does not transcribe the score",
    "source_url": "https://arxiv.org/html/2507.17448v1",
    "date_found": "2026-11-16"
   },
   {
    "field": "release_milestones.weights_release",
    "was": "8B weights released 2026-01-12",
    "now": "First 8B weights repository created 2025-11-20 (announced 2025-11-22); 2026-01-12 was only a README edit; another 8B repository was created 2026-08-20",
    "source_url": "https://huggingface.co/OpenDFM/RetroDFM-R-v0-8B",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2026-08-24 (paper v2)",
    "now": "2026-08-25 (code repository commit, no release object)",
    "source_url": "https://github.com/OpenDFM/RetroDFM-R",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "benchmarks_reported[0]",
   "release_milestones.weights_release",
   "release_milestones.code_release",
   "latest_update"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-retrodfm-r/",
   "en": "/bio/benchmark/models/retrodfm-r/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-rf3",
  "kind": "model",
  "short_slug": "rf3",
  "names": {
   "display": "RF3",
   "full": "RoseTTAFold3",
   "full_null_reason": null,
   "aliases": [
    "RF3",
    "RoseTTAFold3",
    "RoseTTAFold 3"
   ],
   "ambiguous_aliases": [
    "RoseTTAFold"
   ]
  },
  "family": null,
  "family_null_reason": "Single model; design models sharing the repository have their own entries",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Open co-folding model from a protein design institute, released under BSD with the AtomWorks data framework",
  "org": {
   "names": [
    "University of Washington Institute for Protein Design",
    "RosettaCommons"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding (arbitrary biomolecular complexes)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "BSD",
   "code_license": "BSD",
   "license_changed": null,
   "note": "Code, weights and data pipeline under BSD; exact BSD variant not recorded this round"
  },
  "first_release": {
   "date": "2025-08-14",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv preprint v1 (AtomWorks and RF3)",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2025.08.14.670328",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-08-14",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.08.14.670328",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-08-14",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv preprint v1",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.08.14.670328",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-09-15",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Announcement of open-source code in the Foundry repository",
    "timestamp_utc": null,
    "source_url": "https://rosettacommons.org/2025/09/15/atomworks-rf3-in-foundry-now-open-source-on-github/",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2025-09-15",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Weights public with the Foundry open-source announcement",
    "timestamp_utc": null,
    "source_url": "https://rosettacommons.org/2025/09/15/atomworks-rf3-in-foundry-now-open-source-on-github/",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-09-15",
    "precision": "day",
    "event_kind": "product-update",
    "basis": "RosettaCommons announcement",
    "timestamp_utc": null,
    "source_url": "https://rosettacommons.org/2025/09/15/atomworks-rf3-in-foundry-now-open-source-on-github/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-16",
   "precision": "day",
   "event_kind": "code-maintenance-only",
   "basis": "Foundry repository commit log",
   "description": "Foundry repository restructure (shared infrastructure moved into model packages); latest tag remains v0.2.0 (2026-06-05); commits to 2026-11-04 are docs; no RF3 weight change",
   "timestamp_utc": null,
   "source_url": "https://github.com/RosettaCommons/foundry/commits/main",
   "verification": "verified",
   "note": "No 'Foundry 1.0' tag exists"
  },
  "status": "active",
  "status_note": "Shared repository has a release and maintenance commits within 12 months",
  "what_it_does": "Successor to RoseTTAFold All-Atom, retrained on the AtomWorks data framework to predict arbitrary biomolecular complexes with better chirality handling. AtomWorks serves both structure prediction and design and is also used by RFdiffusion2/3. Like OpenFold3 and Protenix it is an open reproduction route, but it comes from a design lab sharing one data pipeline with design models.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB, processed through the AtomWorks pipeline",
    "source_url": "https://doi.org/10.1101/2025.08.14.670328"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "maintainer leaderboard (2024-01+ subset)",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "confidence": "high",
    "verification": "verified",
    "note": "Listed on the maintainer leaderboard"
   }
  ],
  "other_evaluations": [
   {
    "label": "Gap to AlphaFold 3",
    "category": "comparison-vs-baseline-only",
    "detail": "The preprint says it narrows the gap between closed AF3 and open implementations; no values taken this round",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: it narrows the gap between closed AlphaFold 3 and open implementations and improves chirality handling and atom-distance conditioning.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Listed on the FoldBench maintainer leaderboard, run by the board.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The University of Washington Institute for Protein Design (IPD) is an academic protein design institute whose director shared the 2024 Nobel Prize in Chemistry; RF3 was developed by two of its labs with the Microsoft AI for Good lab.",
   "funding": "Predecessor RoseTTAFold All-Atom acknowledged The Audacious Project, HHMI, the Gates Foundation, Open Philanthropy, Schmidt Futures, the Helmsley Trust and a Microsoft donation; RF3's full funding statement was not transcribed.",
   "funding_null_reason": null,
   "products_partnerships": "Product line: RoseTTAFold All-Atom (predecessor), the RFdiffusion series (design) and RF3 (open-weight co-folding), with code in the Foundry repository; collaboration with Microsoft.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-rfdiffusion2-3",
    "model-openfold3",
    "model-protenix",
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "No verifiable self-reported numbers exist, so it cannot be called better or worse than AlphaFold 3 or Boltz.",
   "The point is AtomWorks: a unified data and training pipeline matters more for design models.",
   "Foundry has tags up to v0.2.0 only; the claim that 'Foundry 1.0 was released' is unsupported.",
   "Design-side enzyme active-site results belong to RFdiffusion2/3, not RF3 itself."
  ],
  "caveats": [
   "The date follows bioRxiv (2025-08-14); some second-hand sources say 2025-09.",
   "No PoseBusters result was found in two search rounds.",
   "No journal version and no RF3-specific new weights found."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2025.08.14.670328",
    "title": "AtomWorks and RF3 bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://rosettacommons.org/2025/09/15/atomworks-rf3-in-foundry-now-open-source-on-github/",
    "title": "RosettaCommons: AtomWorks and RF3 open-sourced in Foundry",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/RosettaCommons/foundry/releases",
    "title": "Foundry GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/RosettaCommons/foundry/commits/main",
    "title": "Foundry commit log",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://neurosnap.ai/service/RoseTTAFold3",
    "title": "Third-party service page: RoseTTAFold3 (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-06-05 (v0.2.0)",
    "now": "2026-10-16 (repository restructure commit)",
    "source_url": "https://github.com/RosettaCommons/foundry/commits/main",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "Foundry 1.0 restructure / Foundry 1.0 released",
    "now": "Removed: no Foundry 1.0 tag exists; latest is v0.2.0 (2026-06-05)",
    "source_url": "https://github.com/RosettaCommons/foundry/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "No verified site benchmark pairing",
    "now": "Listed on the FoldBench maintainer leaderboard (official participation)",
    "source_url": "https://github.com/BEAM-Labs/FoldBench",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.code_release",
   "release_milestones.weights_release",
   "open_status.code_license",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-rf3/",
   "en": "/bio/benchmark/models/rf3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-rfantibody",
  "kind": "model",
  "short_slug": "rfantibody",
  "names": {
   "display": "RFantibody",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "RFantibody"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "antibody",
  "domains_secondary": [],
  "one_liner": "De novo antibody design pipeline for specified epitopes: generate backbones, design sequences, filter by structure",
  "org": {
   "names": [
    "University of Washington Institute for Protein Design"
   ],
   "global_entity_ids": []
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "De novo antibody design pipeline",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "MIT licence covers code and weights; no training code provided"
  },
  "first_release": {
   "date": "2024-03-18",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 preprint (single-domain antibody version); the code repository creation date is not used",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.03.14.585103",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-03-18",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.03.14.585103",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-11-05",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Nature online publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41586-025-09721-5",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2026-03-05",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub adds JSON input and TCR-MHC support",
    "timestamp_utc": null,
    "source_url": "https://github.com/RosettaCommons/RFantibody",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-03-05",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub code release adding JSON input and TCR-MHC support",
   "description": "Later items are a docs site (2026-09-23) and a licence clarification covering weights (2026-11-05); whether these count as updates is a policy call",
   "timestamp_utc": null,
   "source_url": "https://github.com/RosettaCommons/RFantibody",
   "verification": "partial",
   "note": "The later docs and licence edits are not folded into this field; marked unverified"
  },
  "status": "active",
  "status_note": "Code release within 12 months",
  "what_it_does": "A fine-tuned diffusion model generates backbones for a specified epitope, a sequence-design tool supplies sequences, and a fine-tuned folding model filters; it covers VHH, scFv and full-length antibody formats.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Diffusion and folding models fine-tuned on public PDB antibody complexes",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Yeast display, SPR, cryo-EM and affinity maturation used for validation only",
    "source_url": null
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Yeast display of VHH designs",
    "modality": "VHH",
    "targets": 4,
    "designs_tested": 9000,
    "hits": null,
    "denominator_basis": "Four epitopes with about 9,000 designs per target; the developers give a 0%–2% success range and 28 VHHs confirmed by SPR; per-target percentages sit in supplementary material and are not cited",
    "success_rate": "success rates range from 0% to 2% across targets; 28 VHHs confirmed by SPR",
    "affinity_best": null,
    "version": "Nature paper",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/",
    "date": null,
    "verification": "verified",
    "note": "The 9,000 figure and the 0%–2% range are verified in the published paper, lifting the earlier unverified flag; developer-run"
   },
   {
    "campaign": "E. coli expression arm",
    "modality": "VHH",
    "targets": 3,
    "designs_tested": 95,
    "hits": null,
    "denominator_basis": "Three further targets with 95 designs each, measured by single-concentration SPR",
    "success_rate": null,
    "affinity_best": null,
    "version": "Nature paper",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/",
    "date": null,
    "verification": "verified",
    "note": "Recorded separately from the yeast-display arm; developer-run"
   },
   {
    "campaign": "Cryo-EM validation",
    "modality": "VHH; scFv",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Structures solved for selected complexes: one VHH complex at 3.0 Å and two scFv structures at 3.6 Å and 6.1 Å",
    "success_rate": null,
    "affinity_best": null,
    "version": "Nature paper",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/",
    "date": null,
    "verification": "verified",
    "note": "Cryo-EM, not negative stain; the institutional news page's looser wording must not replace the paper's resolutions"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Affinity maturation",
    "category": "wet-lab-validation",
    "detail": "Developers report initial binders matured to single-digit nM while keeping epitope selectivity",
    "source_type": "self"
   },
   {
    "label": "scFv and full-length IgG designs",
    "category": "wet-lab-validation",
    "detail": "Developers extend the pipeline to scFv and IgG formats with yeast-display confirmation",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report the first atomically accurate de novo antibody design to a specified epitope, validated by cryo-EM",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A competitor's preprint uses it as a baseline, and an independent compilation shows a low structure-filter pass rate (computational, not wet lab)",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The developer is a nonprofit protein-design institute at the University of Washington",
   "funding": "Funders include NIH, Microsoft, Open Philanthropy, the Gates Foundation and HHMI",
   "funding_null_reason": null,
   "products_partnerships": "Related technology is licensed to an AI-native biotech company; some authors are co-inventors on a patent",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-rfdiffusion2-3",
    "model-rf3",
    "model-iggm",
    "model-tfold",
    "model-origin-1"
   ]
  },
  "reading_notes": [
   "Only the 0%–2% range may be used; per-target percentages in the supplement cannot be cited",
   "The two arms (9,000 and 95 designs per target) differ greatly in scale and must be read separately",
   "No results reported on any of the 44 benchmarks"
  ],
  "caveats": [
   "Whether the docs site and licence edits count as updates is a policy call; marked unverified",
   "All experiments were developer-run, not independent replications"
  ],
  "sources": [
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/",
    "title": "PubMed/PMC record",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.03.14.585103",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1038/s41586-025-09721-5",
    "title": "Crossref record 10.1038/s41586-025-09721-5",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/RosettaCommons/RFantibody",
    "title": "GitHub repository RosettaCommons/RFantibody",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.ipd.uw.edu/2025/11/rfantibody-in-nature",
    "title": "Official page ipd.uw.edu",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "Electron microscopy validation, no resolution given",
    "now": "Cryo-EM at 3.0 Å / 3.6 Å / 6.1 Å",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "The 9,000-per-target figure and 0%–2% range marked unverified",
    "now": "Verified in the published paper and citable; the 95-per-target expression arm is added",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12727541/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-rfantibody/",
   "en": "/bio/benchmark/models/rfantibody/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-rfdiffusion2-3",
  "kind": "model",
  "short_slug": "rfdiffusion2-3",
  "names": {
   "display": "RFdiffusion2 / RFdiffusion3",
   "full": null,
   "full_null_reason": "Two models in one record, so there is no single full name",
   "aliases": [
    "RFdiffusion2",
    "RFdiffusion3",
    "RFD2",
    "RFD3"
   ],
   "ambiguous_aliases": [
    "RFdiffusion"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "RFdiffusion2",
     "id_or_null": null,
     "first_release": {
      "date": "2025-04-10",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "Preprint (journal version online 2025-12-03)",
      "timestamp_utc": null,
      "source_url": "https://doi.org/10.1038/s41592-025-02975-x",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "RFdiffusion3",
     "id_or_null": null,
     "first_release": {
      "date": "2025-09-18",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "Preprint v1 (v2 2025-11-19; open source 2025-12-03)",
      "timestamp_utc": null,
      "source_url": "https://doi.org/10.1101/2025.09.18.676967",
      "verification": "verified",
      "note": null
     }
    }
   ],
   "boundary": "Covers RFD2 and RFD3 only; the antibody fine-tuned line, the 2026-10 latent-reasoning successor branch and the sibling structure predictor are each recorded separately."
  },
  "family_null_reason": null,
  "domain": "protein",
  "domains_secondary": [
   "generation",
   "structure"
  ],
  "one_liner": "Atomic-level diffusion design family: RFD2 scaffolds enzyme active sites, RFD3 designs all-atom interfaces",
  "org": {
   "names": [
    "University of Washington Institute for Protein Design",
    "Rosetta Commons"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Design (diffusion, atomic level)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "RFD3 ships weights and training code through a community foundation's Foundry; licence text was not retrieved this round, so the terms are left empty"
  },
  "first_release": {
   "date": "2025-04-10",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "RFD2 preprint (date carried from the existing note); the RFD3 preprint 2025-09-18 and open-source release 2025-12-03 are listed separately",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1101/2025.09.18.676967",
   "verification": "partial",
   "note": "Unverified: the first release covers RFD2 only; see the family field for the boundary"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-04-10",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "RFD2 preprint (date carried from the existing note); the RFD3 preprint 2025-09-18 and open-source release 2025-12-03 are listed separately",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.09.18.676967",
    "verification": "partial",
    "note": "Unverified: the first release covers RFD2 only; see the family field for the boundary"
   },
   "paper_v1": {
    "date": "2025-04-10",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "RFD2 preprint (date carried from the existing note); the RFD3 preprint 2025-09-18 and open-source release 2025-12-03 are listed separately",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.09.18.676967",
    "verification": "partial",
    "note": "Unverified: the first release covers RFD2 only; see the family field for the boundary"
   },
   "journal_online": {
    "date": "2026-10-15",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "RFD3 journal publication (Crossref issued); the RFD2 journal version went online 2025-12-03",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1126/science.aeh2982",
    "verification": "verified",
    "note": "Corrects the site's claim that RFD3 was still preprint-only as of 2026-11-16"
   },
   "code_release": {
    "date": "2025-12-03",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "RFD3 open-sourced with training code through the community Foundry",
    "timestamp_utc": null,
    "source_url": "https://www.ipd.uw.edu/2025/12/rfdiffusion3-now-available",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-12-03",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "RFD3 weights released with the open-source drop",
    "timestamp_utc": null,
    "source_url": "https://www.ipd.uw.edu/2025/12/rfdiffusion3-now-available",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-12-03",
    "precision": "day",
    "event_kind": "official-announcement",
    "basis": "Institute release page",
    "timestamp_utc": null,
    "source_url": "https://www.ipd.uw.edu/2025/12/rfdiffusion3-now-available",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-15",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "RFD3 journal publication (Crossref issued 2026-10-15)",
   "description": "RFD3's journal version appeared; the Foundry's latest tag and package release is v0.2.0 (2026-06-05) with no 1.0 tag, and commits to 2026-11-04 are untagged.",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1126/science.aeh2982",
   "verification": "verified",
   "note": "The earlier 'Foundry 1.0 released' claim has no such tag and is removed"
  },
  "status": "active",
  "status_note": "Journal publication, a tagged release and ongoing commits within 12 months",
  "what_it_does": "RFD2 generates enzyme backbones conditioned on catalytic-residue geometry; RFD3 moves to a co-folding-style architecture and generates all-atom designs that interact with proteins, DNA and small molecules, with self-reported lower compute cost. Unlike the backpropagation route of BindCraft it is a trained generative model, and compared with BoltzGen it leans toward enzymes and all-atom interfaces.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Distilled PDB and structure-database data",
    "source_url": "https://doi.org/10.1101/2025.09.18.676967"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data",
    "source_url": "https://doi.org/10.1101/2025.09.18.676967"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No own training data",
    "source_url": "https://doi.org/10.1101/2025.09.18.676967"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Yes: small-scale enzyme and DNA-binding-protein validation",
    "source_url": "https://doi.org/10.1038/s41592-025-02975-x"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "RFD2 enzyme design (self-reported)",
    "modality": "enzyme",
    "targets": 41,
    "designs_tested": 96,
    "hits": null,
    "denominator_basis": "Backbones were generated for all 41 active sites (in silico; the comparator managed 16), and for each of three catalytic mechanisms an active candidate appeared within fewer than 96 tested sequences",
    "success_rate": null,
    "affinity_best": null,
    "version": "journal version",
    "source_type": "self",
    "source_url": "https://doi.org/10.1038/s41592-025-02975-x",
    "date": {
     "date": "2025-12-03",
     "precision": "day",
     "event_kind": "journal-online",
     "basis": "RFD2 journal version online",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1038/s41592-025-02975-x",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "Verified at abstract level only, without reading full-text kinetics; scaffolding is not activity"
   },
   {
    "campaign": "RFD3 wet-lab characterisation (self-reported)",
    "modality": "DNA-binding protein; hydrolase",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "The preprint describes characterisation of a DNA-binding protein and a cysteine hydrolase at abstract level only, with no counts",
    "success_rate": null,
    "affinity_best": null,
    "version": "preprint v2",
    "source_type": "self",
    "source_url": "https://doi.org/10.1101/2025.09.18.676967",
    "date": {
     "date": "2025-11-19",
     "precision": "day",
     "event_kind": "paper-revision",
     "basis": "Preprint v2",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1101/2025.09.18.676967",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "Verified at abstract level with no figures published; developer-run, not independent replication"
   },
   {
    "campaign": "Successor-branch enzyme test (press release)",
    "modality": "enzyme",
    "targets": 1,
    "designs_tested": 67,
    "hits": 4,
    "denominator_basis": "67 designs were tested; this belongs to the 2026-10 latent-reasoning successor branch, not to RFD2 or RFD3 themselves",
    "success_rate": "4 of 67 designs showed clear activity, the best converting 79% of the substrate amine",
    "affinity_best": null,
    "version": "institutional press release",
    "source_type": "partner-lab",
    "source_url": "https://www.manchester.ac.uk/about/news/ai-designed-enzyme-for-drug-synthesis/",
    "date": {
     "date": "2026-10-22",
     "precision": "day",
     "event_kind": "results-published",
     "basis": "Partner university news release",
     "timestamp_utc": null,
     "source_url": "https://www.manchester.ac.uk/about/news/ai-designed-enzyme-for-drug-synthesis/",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "note": "Not peer reviewed, and the activity is catalytic conversion rather than binding; partner-lab executed, not independent replication"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-40-motifbench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "official board entry (RFdiffusion3, 2026-08 snapshot)",
    "source_url": "https://github.com/blt2114/MotifBench",
    "confidence": "high",
    "verification": "verified",
    "note": "RFD3 is an official entry on that board; the Chinese model page's 'no entry found' was wrong and is corrected here, with no score quoted"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own 41-active-site set",
    "category": "own-task-set",
    "detail": "An author-built set of enzyme active sites reporting how many could be scaffolded versus the comparator (self-reported)",
    "source_type": "self"
   },
   {
    "label": "Metallohydrolase design",
    "category": "wet-lab-validation",
    "detail": "RFD2-designed metallohydrolases were tested in two rounds and the crystal structures match the design models (journal paper)",
    "source_type": "self"
   },
   {
    "label": "Successor-branch sampling-efficiency comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "A vendor blog claims the successor branch needs fewer sampling steps and less memory; not peer reviewed and unverified",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report that RFD3 beats prior methods across in silico evaluations at about one tenth of the compute cost, with wet-lab characterisation of a DNA-binding protein and a hydrolase (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The MotifBench board lists an RFD3 entry (no score quoted here).",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "A protein design institute at a US public university, funded by several foundations, a large software company and NIH; RFD3 is released through a community foundation's open-source channel rather than run by one company.",
   "funding": null,
   "funding_null_reason": "Institutionally funded; no funding rounds at model level",
   "products_partnerships": "It succeeds the institute's structure predictor and earlier diffusion design model; design companies spun out of the institute are not tied to this model as products.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-rf3",
    "model-rfantibody",
    "model-proteina-complexa",
    "model-bindcraft",
    "model-boltzgen"
   ]
  },
  "reading_notes": [
   "The active-site set is author-built, and scaffolding a backbone is not the same as catalytic activity.",
   "The wet-lab volume is small and cannot support a claim that enzyme design is mature.",
   "RFD3 is open and ships training code, making it one of the few reproducible all-atom design routes.",
   "MotifBench is its only site board relation; every other evaluation sits on author-built sets."
  ],
  "caveats": [
   "One record covers two models, so RFD2 and RFD3 must be told apart when reading figures.",
   "The successor branch's enzyme test comes only from an institutional press release and is not peer reviewed.",
   "The earlier 'Foundry 1.0 released' claim has no such tag and is removed; the latest tag is v0.2.0 (2026-06-05).",
   "The first-release date covers RFD2 only and is marked unverified."
  ],
  "sources": [
   {
    "url": "https://www.ipd.uw.edu/2025/12/rfdiffusion3-now-available",
    "title": "Institute release page: RFD3 now available",
    "kind": "official-site",
    "date": {
     "date": "2025-12-03",
     "precision": "day",
     "event_kind": "official-announcement",
     "basis": "Release page published",
     "timestamp_utc": null,
     "source_url": "https://www.ipd.uw.edu/2025/12/rfdiffusion3-now-available",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://doi.org/10.1101/2025.09.18.676967",
    "title": "RFD3 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41592-025-02975-x",
    "title": "RFD2 journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1126/science.aeh2982",
    "title": "RFD3 journal record (Crossref)",
    "kind": "archive",
    "date": null
   },
   {
    "url": "https://github.com/RosettaCommons/foundry",
    "title": "Foundry repository (tags and versions)",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/blt2114/MotifBench",
    "title": "MotifBench official repository board",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://www.manchester.ac.uk/about/news/ai-designed-enzyme-for-drug-synthesis/",
    "title": "Partner university news release (successor branch)",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.description",
    "was": "The record stated that Foundry 1.0 had been released",
    "now": "No such tag exists and the claim is removed; the latest tag and package release is v0.2.0 (2026-06-05)",
    "source_url": "https://github.com/RosettaCommons/foundry",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "The Chinese model page said no MotifBench entry was found",
    "now": "RFD3 is an official entry on that board (2026-08 snapshot) and the relation is official participation",
    "source_url": "https://github.com/blt2114/MotifBench",
    "date_found": "2026-11-16"
   },
   {
    "field": "release_milestones.journal_online",
    "was": "The site said RFD3 was still preprint-only as of 2026-11-16",
    "now": "RFD3's journal version appeared 2026-10-15 and is listed as its own milestone",
    "source_url": "https://api.crossref.org/works/10.1126/science.aeh2982",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "other_evaluations[2]",
   "open_status.code_license",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-rfdiffusion2-3/",
   "en": "/bio/benchmark/models/rfdiffusion2-3/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-ribospan",
  "kind": "model",
  "short_slug": "ribospan",
  "names": {
   "display": "RIBOSPAN",
   "full": null,
   "aliases": [
    "RIBOSPAN",
    "RIBOSPAN-FM"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single model whose four checkpoints are configuration axes inside one record",
  "domain": "genomics-rna",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "1.61B-parameter bidirectional RNA foundation model with a native 10,240-nt context and single-nucleotide tokens, able to take a full transcript at once",
  "org": {
   "names": [
    "Shanghai Innovation Institute",
    "CAS Center for Excellence in Molecular Cell Science",
    "Shanghai Jiao Tong University"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "RNA foundation model (representation)",
  "open_status": {
   "code": "open",
   "weights": "non-commercial",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code is public; weights carry the developer's own non-commercial research licence, with one checkpoint gated and another still marked as coming"
  },
  "first_release": {
   "date": "2026-08-24",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1",
   "event_kind": "preprint-posted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2608.22849",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-08-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2608.22849",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-08-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2608.22849",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-08-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Public code repository",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/GAIR-NLP/RIBOSPAN-FM",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2026-09-01",
    "precision": "day",
    "year_hint": null,
    "basis": "Model hub collection updated: short-context checkpoints public, one long-context checkpoint gated and another marked as coming",
    "event_kind": "weights-release",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/SII-GAIR-NLP/RIBOSPAN-FM",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-09-01",
   "precision": "day",
   "year_hint": null,
   "basis": "Model hub collection update (weights axis)",
   "event_kind": "new-weights",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/SII-GAIR-NLP/RIBOSPAN-FM",
   "verification": "partial",
   "description": "After the update one long-context checkpoint is still marked as coming; the preprint also has a v2 revision dated 2026-08-28, whose date was not independently re-checked this round",
   "note": null
  },
  "status": "active",
  "status_note": "Preprint and weights released within 12 months",
  "what_it_does": "Addresses full-length mRNA exceeding existing RNA models' context using dense bidirectional attention, single-nucleotide tokens and attention-isolated sequence packing, pretraining at a native 10,240-nt context so a whole transcript fits at once; the same backbone also carries a discrete-diffusion mRNA generation line whose evaluation is not covered on this site. Comparators are several encoder models of the same class, with a large version of Evo 2 as a decoder-side reference.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "67.6 million deduplicated RNA sequences and 85.7 billion nucleotide tokens from specified versions of three public sequence and genome-annotation resources",
    "source_url": "https://huggingface.co/SII-GAIR-NLP/RIBOSPAN-FM"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2608.22849"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2608.22849"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Downstream evaluation uses existing experimental datasets; no new experiments",
    "source_url": "https://arxiv.org/abs/2608.22849"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-39-rnagym",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/pdf/2608.22849",
    "confidence": "high",
    "second_hand": false,
    "note": "Zero-shot mutational fitness, self-reported for all four checkpoints under that evaluation's per-class averaging; the decoder comparator's values are taken from the evaluation's own report rather than re-run"
   }
  ],
  "other_evaluations": [
   {
    "label": "Whole-transcript linear-probing benchmark",
    "category": "public-benchmark-not-on-site",
    "detail": "Self-reported results on six task classes of an mRNA benchmark not among this site's 44, against encoder models of the same class",
    "source_type": "self"
   },
   {
    "label": "Own long-context reconstruction and representation tests",
    "category": "own-task-set",
    "detail": "Long-context reconstruction and RNA-class separation under frozen representations, defined by the authors",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report the strongest representations among encoder models of its class and the ability to process a whole transcript at once in a native long context.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party reproduction was found; all evaluation results are self-reported",
  "background": {
   "org_profile": "The lead organisation is a new AI research and doctoral-training institute founded in Shanghai in 2024-09, not a company; collaborators include a national molecular cell science centre and a university.",
   "funding": "No separate funding disclosure was found; backers and scale are not public.",
   "products_partnerships": "The lead team was previously noted for work on reasoning in large models, and this model is its entry into RNA; no pharmaceutical partnership is disclosed.",
   "related_models": [
    "model-evo-2",
    "model-rna-structure-models"
   ]
  },
  "reading_notes": [
   "Its lead is confined to encoder models of its own class; the decoder-side reference scores higher on the same evaluation.",
   "Short-context checkpoints do better on some tasks, with the long-context advantage concentrated where a whole transcript is needed.",
   "All results are self-reported with checkpoint choice and protocol set by the authors, and no third party has reproduced them.",
   "A non-commercial licence and gated checkpoints limit external use and checking."
  ],
  "caveats": [
   "The decoder comparator's values come from the evaluation's own report, not a re-run in the same environment.",
   "One long-context checkpoint was still unreleased at the check date, so its reported results cannot be reproduced.",
   "The other whole-transcript benchmark is not among this site's 44, so its values cannot be listed beside entries here."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2608.22849",
    "title": "Preprint",
    "kind": "preprint",
    "date": "2026-08-24"
   },
   {
    "url": "https://arxiv.org/pdf/2608.22849",
    "title": "Preprint full text PDF",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/SII-GAIR-NLP/RIBOSPAN-FM",
    "title": "Model hub collection",
    "kind": "model-hub",
    "date": null
   },
   {
    "url": "https://github.com/GAIR-NLP/RIBOSPAN-FM",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "latest_update",
   "release_milestones.weights_release",
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-ribospan/",
   "en": "/bio/benchmark/models/ribospan/"
  },
  "names_full_null_reason": "The name is a model name; the paper gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-rna-structure-models",
  "kind": "model",
  "short_slug": "rna-structure-models",
  "names": {
   "display": "RhoFold+/DRfold2/trRosettaRNA",
   "full": null,
   "full_null_reason": "Family entry; no single expanded name",
   "aliases": [
    "RhoFold+",
    "DRfold2",
    "trRosettaRNA",
    "trRosettaRNA2"
   ],
   "ambiguous_aliases": [
    "RhoFold",
    "DRfold",
    "trRosetta"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "RhoFold+",
     "id_or_null": null,
     "first_release": {
      "date": "2022-07-04",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "arXiv v1 (earliest family member; unverified)",
      "timestamp_utc": null,
      "source_url": "https://www.nature.com/articles/s41592-024-02487-0",
      "verification": "unverified",
      "note": null
     }
    },
    {
     "name": "DRfold2",
     "id_or_null": null,
     "first_release": {
      "date": "2025-03-11",
      "precision": "day",
      "event_kind": "preprint-posted",
      "basis": "bioRxiv v1",
      "timestamp_utc": null,
      "source_url": "https://www.biorxiv.org/content/10.1101/2025.03.05.641632v1.full",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "trRosettaRNA2",
     "id_or_null": null,
     "first_release": {
      "date": "2025-04",
      "precision": "month",
      "event_kind": "preprint-posted",
      "basis": "bioRxiv preprint (date unverified)",
      "timestamp_utc": null,
      "source_url": "https://doi.org/10.1101/2025.04.09.647915",
      "verification": "unverified",
      "note": null
     }
    }
   ],
   "boundary": "Covers only these three RNA 3D prediction lines, including the earlier trRosettaRNA; other RNA models such as RNAPro are separate entries."
  },
  "family_null_reason": null,
  "domain": "structure",
  "domains_secondary": [
   "genomics-rna"
  ],
  "one_liner": "Three independent RNA 3D prediction lines from three academic groups, not mutually comparable",
  "org": {
   "names": [
    "CUHK / Harvard / MIT collaboration",
    "National University of Singapore group",
    "Shandong University (Qingdao) group"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN",
   "SG",
   "US"
  ],
  "model_class": "RNA 3D structure prediction",
  "open_status": {
   "code": "partial",
   "weights": "research-only",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "RhoFold+ ships weights and inference code under Apache-2.0, the DRfold2 repository states no licence, and trRosettaRNA2 offers only a free web server"
  },
  "first_release": {
   "date": "2022-07-04",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "Preprint v1 of the earliest member, RhoFold+ (member and date not re-verified, unverified)",
   "timestamp_utc": null,
   "source_url": "https://www.nature.com/articles/s41592-024-02487-0",
   "verification": "unverified",
   "note": "Family entry: the first release is the earliest member's, not a joint release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2022-07-04",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "The earliest member's preprint v1 (unverified)",
    "timestamp_utc": null,
    "source_url": "https://www.nature.com/articles/s41592-024-02487-0",
    "verification": "unverified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-03-11",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "DRfold2 bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.03.05.641632v1.full",
    "verification": "partial",
    "note": null
   },
   "journal_online": {
    "date": "2024-11-21",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "RhoFold+ journal publication",
    "timestamp_utc": null,
    "source_url": "https://www.nature.com/articles/s41592-024-02487-0",
    "verification": "partial",
    "note": null
   },
   "code_release": {
    "date": null,
    "precision": null,
    "event_kind": "code-release",
    "basis": "No sourced first code-release date for any of the three lines",
    "timestamp_utc": null,
    "source_url": null,
    "verification": "unverified",
    "note": "Repository creation dates are not used as evidence"
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-04-21",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "trRosettaRNA2's journal publication (DOI and date second-hand, Crossref rate-limited)",
   "description": "trRosettaRNA2 appeared in a journal; DRfold2 was published 2026-02-17 and RhoFold+ on 2024-11-21",
   "timestamp_utc": null,
   "source_url": "https://yanglab.qd.sdu.edu.cn/papers/Wang_NP_2026.pdf",
   "verification": "partial",
   "note": "A member event inside the family, not a joint update of all three lines (unverified)"
  },
  "status": "maintained",
  "status_note": "All three published in journals between 2024 and 2026; repositories are in maintenance",
  "what_it_does": "The three lines predict RNA 3D structure with, respectively, an RNA language model plus a structure module, a compound language model plus denoising, and a Transformer that predicts geometric restraints before energy minimisation. With only a few thousand experimental RNA structures, all three lean on sequence databases and language models, and absolute accuracy stays well below protein co-folding such as AlphaFold3.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "A few thousand RNA structures in the PDB plus sequence databases such as RNAcentral; the RhoFold+ language model was pretrained on about 23.7M sequences",
    "source_url": "https://www.nature.com/articles/s41592-024-02487-0"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "CASP16 automated-server section (trRosettaRNA2); CASP15 retrospective (RhoFold+)",
    "source_url": "https://predictioncenter.org/casp16/doc/presentations/Day-3/Day3-06-Wang_CASP16_Yang-Server.pdf",
    "confidence": "medium",
    "verification": "partial",
    "note": "Family cell: only trRosettaRNA2 took part in CASP16, while RhoFold+'s CASP15 evaluation is retrospective"
   },
   {
    "benchmark_id": "bench-25-rnapuzzles",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "single-chain RNA-Puzzles targets, retrospective (RhoFold+)",
    "source_url": "https://www.nature.com/articles/s41592-024-02487-0",
    "confidence": "medium",
    "verification": "partial",
    "note": "The RNA-Puzzles results of DRfold2 and trRosettaRNA2 are unconfirmed: the former pools targets with other sets and the latter's results belong to its predecessor (unverified)"
   }
  ],
  "other_evaluations": [
   {
    "label": "Self-built RNA test set",
    "category": "self-report",
    "detail": "The DRfold2 preprint compares against a co-folding model on 28 self-assembled structures, excluding targets over a length cut-off (self-reported)",
    "source_type": "self"
   },
   {
    "label": "Evidence from a public challenge",
    "category": "third-party-evaluation",
    "detail": "In part one of an RNA 3D folding challenge the top three tied with the previous blind-test winner within statistical error, and the best strategy was template retrieval without deep learning (abstract, second-hand)",
    "source_type": "third"
   },
   {
    "label": "Review comparison",
    "category": "third-party-evaluation",
    "detail": "A review compares six RNA structure methods on one set of data; the ranking itself was not obtained (third party)",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report each line beating contemporaneous methods on retrospective RNA sets; the protocols differ and are not comparable inside the family (self-reported).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The CASP16 organisers' assessment of the automated-server section is the firmest third-party evidence, and it belongs to the trRosettaRNA2 line alone.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Three academic groups and no company: a CUHK collaboration with Harvard and MIT (RhoFold+), a group at the National University of Singapore (DRfold2), and a group at Shandong University in Qingdao (trRosettaRNA).",
   "funding": null,
   "funding_null_reason": "Academic projects; funding amounts not public",
   "products_partnerships": "RhoFold+ releases weights and inference code, the DRfold2 repository states no licence, and the trRosettaRNA2 site says its server is free for all users; the three compete with each other.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-rnapro",
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "The three lines use inconsistent RNA-Puzzles protocols and two are unconfirmed, so do not rank them on the page.",
   "The CASP16 server-section placing is the firmest evidence, but only for trRosettaRNA2.",
   "Template retrieval remains very strong on RNA; deep learning's edge on template-free targets is unproven.",
   "The family's first-release date is the earliest member's and has not been re-verified."
  ],
  "caveats": [
   "The first-release member and date were not re-verified and are shown as unverified.",
   "The latest update is one member's journal publication, not a joint update.",
   "Experimental RNA structures are scarce, so absolute accuracy is low under every protocol.",
   "Openness differs across the three; only one line states a clear licence."
  ],
  "sources": [
   {
    "url": "https://www.nature.com/articles/s41592-024-02487-0",
    "title": "RhoFold+ journal paper",
    "kind": "journal",
    "date": {
     "date": "2024-11-21",
     "precision": "day",
     "event_kind": "journal-publication",
     "basis": "Journal publication",
     "timestamp_utc": null,
     "source_url": "https://www.nature.com/articles/s41592-024-02487-0",
     "verification": "partial",
     "note": null
    }
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.03.05.641632v1.full",
    "title": "DRfold2 preprint full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2025.04.09.647915",
    "title": "trRosettaRNA2 preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1371/journal.pbio.3003659",
    "title": "DRfold2 journal publication",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://predictioncenter.org/casp16/doc/presentations/Day-3/Day3-06-Wang_CASP16_Yang-Server.pdf",
    "title": "CASP16 server-section presentation",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update.description",
    "was": "DRfold2 existed only as a preprint",
    "now": "DRfold2 was published in a journal on 2026-02-17",
    "source_url": "https://doi.org/10.1371/journal.pbio.3003659",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported.bench-25-rnapuzzles",
    "was": "All three lines had RNA-Puzzles results",
    "now": "Only the RhoFold+ retrospective holds; the other two are unconfirmed (unverified)",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.03.05.641632v1.full",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release.date",
   "family.members[0].first_release.date",
   "family.members[2].first_release.date",
   "latest_update.date",
   "benchmarks_reported[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-rna-structure-models/",
   "en": "/bio/benchmark/models/rna-structure-models/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-rnapro",
  "kind": "model",
  "short_slug": "rnapro",
  "names": {
   "display": "RNAPro",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "RNAPro"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "structure",
  "domains_secondary": [
   "genomics-rna"
  ],
  "one_liner": "Open model predicting all-atom RNA 3D structure, about 488M parameters",
  "org": {
   "names": [
    "NVIDIA",
    "Stanford University"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": "Joint work by a big-tech lab and an academic lab",
  "country": [
   "US"
  ],
  "model_class": "RNA structure prediction",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "NVIDIA Open Model License",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "Code Apache-2.0; weights under the NVIDIA Open Model License, commercial use allowed"
  },
  "first_release": {
   "date": "2026-01-09",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "First commits on GitHub and Hugging Face on the same day",
   "timestamp_utc": null,
   "source_url": "https://github.com/NVIDIA-Digital-Bio/RNAPro",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2026-01-09",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "First GitHub commit",
    "timestamp_utc": null,
    "source_url": "https://github.com/NVIDIA-Digital-Bio/RNAPro",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-01-09",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "First Hugging Face commit",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/nvidia/RNAPro-Public-Best-500M/commits/main",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-06-05",
   "precision": "day",
   "event_kind": "code-maintenance-only",
   "basis": "README documentation update (post-processing guide, citation, data provenance)",
   "description": "Documentation only; no new weights or version",
   "timestamp_utc": null,
   "source_url": "https://api.github.com/repos/NVIDIA-Digital-Bio/RNAPro/commits?per_page=8",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Documentation maintenance only since release",
  "what_it_does": "Predicts all-atom RNA 3D structure by combining a co-folding architecture, RibonanzaNet2, MSA and template modeling at about 488M parameters; grew out of a public blind code competition entry.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Protenix training set plus public RNA folding competition data (5,135 sequences with all-atom labels)",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "Evaluation used a private set of newly solved structures",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-25-rnapuzzles",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "RNA-Puzzles subset within a competition private leaderboard",
    "source_url": "https://www.biorxiv.org/content/10.64898/2025.12.30.696949.full.pdf",
    "confidence": "medium",
    "verification": "partial",
    "note": "Developer-reported on an RNA-Puzzles subset inside a competition private leaderboard; protocol differs from the official one"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "NVIDIA is a listed GPU and AI company; RNAPro belongs to its open model family and is internally funded; the academic side is a Stanford biochemistry team",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "Builds on Protenix and RibonanzaNet2; grew out of a blind competition with 1,700+ teams",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-protenix",
    "model-rna-structure-models"
   ]
  },
  "reading_notes": [
   "Its only pairing is a subset inside a competition private leaderboard, not comparable with the official contest",
   "The weights licence allows commercial use, which is uncommon among comparable RNA structure models"
  ],
  "caveats": [
   "Preprint posting date not verified",
   "The evaluation set is private newly solved structures and cannot be reproduced externally"
  ],
  "sources": [
   {
    "url": "https://api.github.com/repos/NVIDIA-Digital-Bio/RNAPro/commits?per_page=8",
    "title": "GitHub repository repos/NVIDIA-Digital-Bio",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/nvidia/RNAPro-Public-Best-500M/commits/main",
    "title": "Hugging Face page nvidia/RNAPro-Public-Best-500M",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/NVIDIA-Digital-Bio/RNAPro",
    "title": "GitHub repository NVIDIA-Digital-Bio/RNAPro",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.64898/2025.12.30.696949.full.pdf",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-rnapro/",
   "en": "/bio/benchmark/models/rnapro/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-robin",
  "kind": "model",
  "short_slug": "robin",
  "names": {
   "display": "Robin",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [],
   "ambiguous_aliases": [
    "Robin"
   ]
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "text-agents",
  "domains_secondary": [],
  "one_liner": "FutureHouse multi-agent discovery system linking literature review, candidate proposal and data analysis",
  "org": {
   "names": [
    "FutureHouse"
   ],
   "global_entity_ids": [
    "futurehouse"
   ]
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Multi-agent discovery system",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code released, but running it needs a FutureHouse platform key"
  },
  "first_release": {
   "date": "2025-05-19",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2505.13400 v1 submitted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2505.13400",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-05-19",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2505.13400",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-05-19",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Nature publication",
    "timestamp_utc": null,
    "source_url": "https://www.nature.com/articles/s41586-026-10652-y",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-05-19",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Nature journal publication",
   "description": "Paper published in Nature",
   "timestamp_utc": null,
   "source_url": "https://www.nature.com/articles/s41586-026-10652-y",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication within 12 months",
  "what_it_does": "Three agents (Crow, Falcon, Finch) handle literature review, candidate proposal and experimental data analysis; demonstrated on dry age-related macular degeneration, with proposed candidates tested in the wet lab.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Base model and retrieval tools use public literature",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Cell phagocytosis assays tested candidates",
    "source_url": "https://arxiv.org/html/2505.13400"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Dry-AMD lab-in-the-loop (two rounds)",
    "modality": "small molecule",
    "targets": null,
    "designs_tested": 10,
    "hits": null,
    "denominator_basis": "Round 1 tested 5 of 30 proposed candidates, round 2 tested 10; no hit rate reported",
    "success_rate": null,
    "affinity_best": null,
    "version": "paper",
    "source_type": "self",
    "source_url": "https://arxiv.org/html/2505.13400",
    "date": null,
    "verification": "verified",
    "note": "Team-run experiments and analysis, not independent; agent and human estimates of the same effect differ (7.5x vs 1.75x); authors call for more dose/time testing"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Dry-AMD lab-in-the-loop",
    "category": "wet-lab-validation",
    "detail": "Two rounds of cell assays: 5 of 30 candidates, then 10",
    "source_type": "self"
   },
   {
    "label": "Effect size conflict",
    "category": "case-study",
    "detail": "Same effect: agent analysis 7.5x, human 1.75x; authors call for more testing",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report the first multi-agent system to autonomously hypothesize, plan, analyze and propose a validated new candidate",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evidence found",
  "background": {
   "org_profile": "FutureHouse is a nonprofit AI-science lab founded in San Francisco in 2023-09; its commercial arm is Edison Scientific",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "Done with Oxford collaborators; related products Kosmos and Finch",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-kosmos",
    "model-biomni"
   ]
  },
  "reading_notes": [
   "Wet-lab work is small and team-run: a demonstration",
   "Agent and human effect sizes differ about 4-fold; both should be shown"
  ],
  "caveats": [
   "No results reported on any of the 44 benchmarks",
   "Round-2 hit outcomes not reported"
  ],
  "sources": [
   {
    "url": "https://api.crossref.org/works/10.1038/s41586-026-10652-y",
    "title": "Crossref record 10.1038/s41586-026-10652-y",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://futurehouse.org/research-announcements/demonstrating-end-to-end-scientific-discovery-with-robin-a-multi-agent-system",
    "title": "Official page futurehouse.org",
    "kind": "company-blog",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2505.13400",
    "title": "arXiv preprint 2505.13400",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.nature.com/articles/s41586-026-10652-y",
    "title": "Journal or proceedings page",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-robin/",
   "en": "/bio/benchmark/models/robin/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-saprot",
  "kind": "model",
  "short_slug": "saprot",
  "names": {
   "display": "SaProt",
   "full": null,
   "full_null_reason": "Short name only; neither paper nor repository gives an expanded name",
   "aliases": [
    "SaProt",
    "SaProt 650M",
    "SaProt 35M",
    "SaProt 1.3B"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "35M and 650M are two sizes of one model, not family members",
  "domain": "protein",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Structure-aware protein language model pairing structure tokens with amino acids",
  "org": {
   "names": [
    "Westlake University",
    "Westlake Representation Learning Lab"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Protein language model (structure-aware)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code and weights public on GitHub and Hugging Face; licence text not retrieved this round, so terms are left empty"
  },
  "first_release": {
   "date": "2023-10-02",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "First public bioRxiv preprint",
   "timestamp_utc": null,
   "source_url": "https://github.com/westlake-repl/SaProt",
   "verification": "partial",
   "note": "The preprint DOI is not Crossref-confirmed; the date comes from the same-day weights repository and README notes (unverified)"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-10-02",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint posted",
    "timestamp_utc": null,
    "source_url": "https://github.com/westlake-repl/SaProt",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2023-10-02",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://github.com/westlake-repl/SaProt",
    "verification": "partial",
    "note": null
   },
   "journal_online": {
    "date": "2025-10-24",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "SaProt/SaprotHub published in Nature Biotechnology",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41587-025-02859-7",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": {
    "date": "2024-12-09",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "1.3B weights noted in the README news",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/westlake-repl/SaProt_650M_AF2",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-10-24",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Journal publication in Nature Biotechnology",
   "description": "SaProt/SaprotHub published in a journal; 1.3B weights earlier on 2024-12-09, README touched in 2026-03",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41587-025-02859-7",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Journal publication and repository upkeep, no new generation",
  "what_it_does": "Combines each residue's amino acid with a local structure token into a structure-aware vocabulary, pretrained on a public predicted-structure database, for zero-shot mutation-effect scoring and fine-tuning, with no extra MSA. It is the simpler predecessor of structure-enhanced zero-shot models such as ProSST and VenusREM.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 40M public predicted structures and sequences, encoded into structure tokens with a structure-alignment tool",
    "source_url": "https://github.com/westlake-repl/SaProt"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data reported",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "zero-shot substitution DMS summary",
    "source_url": "https://github.com/OATML-Markslab/ProteinGym",
    "confidence": "high",
    "verification": "verified",
    "note": "Official board entry; the repository notes it once led in 2024-04 and no longer sits at the top, with position moving between versions"
   },
   {
    "benchmark_id": "bench-26-flip",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "Meltome human-cell (called Thermostability in the paper); AAV 2-vs-rest",
    "source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/file/1c42513b8895ab11fbbb5b7e8e6b6b02-Paper-Conference.pdf",
    "confidence": "high",
    "verification": "verified",
    "note": "Developer report on the official splits; a separate third-party re-evaluation on the GB1 extrapolation split disagrees"
   },
   {
    "benchmark_id": "bench-34-abbibench",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "antibody binding-affinity correlation, structure-aware LM category",
    "source_url": null,
    "confidence": "medium",
    "verification": "unverified",
    "note": "This pairing rests only on the site's own page; no primary document was opened (unverified)"
   }
  ],
  "other_evaluations": [
   {
    "label": "Third-party FLIP GB1 extrapolation re-evaluation",
    "category": "third-party-re-evaluation",
    "detail": "A third-party paper re-ran the low-vs-high split and found it no better than a single-sequence baseline (figures not quoted here)",
    "source_type": "third"
   },
   {
    "label": "Time-dependence of the board position",
    "category": "public-benchmark-not-on-site",
    "detail": "A maintainer blog post discusses it as an example of the structure-token route when weighing scale against gains",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report that adding structure tokens to the vocabulary lets a mid-sized model beat same-size sequence-only models on zero-shot mutation effects without extra MSA.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A third-party paper independently re-ran it on a FLIP extrapolation split and did not reproduce the advantage; no wet lab, no independent replication.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "An academic lab with no separate company: a representation-learning lab at Westlake University in Hangzhou working on AI for life science and on deep learning for retrieval and recommendation. The model appeared at a 2024 machine-learning conference and in a journal in 2025.",
   "funding": null,
   "funding_null_reason": "Paper acknowledgements not retrieved; funding left blank",
   "products_partnerships": "A companion hosting and fine-tuning platform (SaprotHub) shipped with the journal version; the structure-token approach is treated as a peer baseline by ProSST and VenusREM, and third-party inference platforms offer it online.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-prosst",
    "model-venusrem",
    "model-esm3",
    "model-esm-c"
   ]
  },
  "reading_notes": [
   "The \"once first\" claim is a 2024 snapshot; board position shifts with versions and newcomers and is not a current statement.",
   "Structure inputs come from public predicted structures, so regions without structure or with low confidence are weak spots.",
   "A third-party re-evaluation on an extrapolation split shows structure tokens do not guarantee better extrapolation.",
   "Computational evaluation only, no wet lab: it speaks to ranking, not to designability."
  ],
  "caveats": [
   "The AbBiBench pairing rests only on this site's page (unverified).",
   "First-release evidence is thin, taken from the same-day weights repository and README (unverified).",
   "No wet-lab validation, and no independent replication of its board position."
  ],
  "sources": [
   {
    "url": "https://proceedings.iclr.cc/paper_files/paper/2024/file/1c42513b8895ab11fbbb5b7e8e6b6b02-Paper-Conference.pdf",
    "title": "SaProt conference paper (2024)",
    "kind": "paper",
    "date": null
   },
   {
    "url": "https://github.com/westlake-repl/SaProt",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41587-025-02859-7",
    "title": "SaProt/SaprotHub journal version",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://huggingface.co/westlake-repl/SaProt_650M_AF2",
    "title": "Weights repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym",
    "title": "ProteinGym official repository and board",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2501.18223v1",
    "title": "Third-party FLIP re-evaluation paper",
    "kind": "third-party-paper",
    "date": null
   },
   {
    "url": "https://pascalnotin.substack.com/p/have-we-hit-the-scaling-wall-for",
    "title": "Board maintainer blog post (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release",
   "benchmarks_reported[2]",
   "release_milestones.weights_release",
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-saprot/",
   "en": "/bio/benchmark/models/saprot/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-saturn",
  "kind": "model",
  "short_slug": "saturn",
  "names": {
   "display": "Saturn",
   "full": null,
   "aliases": [
    "Saturn"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "The predecessor algorithm and downstream synthesizability work have no entity records of their own",
  "domain": "small-molecule",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Generative molecular design model aimed at sample-efficient search under a limited oracle budget",
  "org": {
   "names": [
    "EPFL (Laboratory of Artificial Chemical Intelligence)"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CH"
  ],
  "model_class": "Molecular generation with reinforcement-learning optimisation",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code is public on GitHub; licence text and whether weights are released separately were not retrieved this round"
  },
  "first_release": {
   "date": "2024-05-27",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2405.17066 v1 submission date",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2405.17066",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-05-27",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2405.17066",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-05-27",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2405.17066",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-03-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal version online (registry-recorded created/issued date)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s42256-026-01200-4",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-05-29",
    "precision": "day",
    "year_hint": null,
    "basis": "News item from the university AI centre",
    "event_kind": "documentation-update",
    "timestamp_utc": null,
    "source_url": "https://ai.epfl.ch/two-ai-platforms-transform-molecular-design/",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-03-17",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1038/s42256-026-01200-4",
   "verification": "verified",
   "description": "Journal version published (DOI 10.1038/s42256-026-01200-4); afterwards the repository merged reaction-condition constraints (2026-06-17) and documentation commits (2026-10-26), counted as secondary code activity",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication and repository commits within 12 months",
  "what_it_does": "Combines a state-space (Mamba) sequence model with a replay-plus-augmentation reinforcement-learning algorithm to optimise molecules when calls to an expensive oracle (docking, quantum calculations) are budget-limited, aiming to satisfy multi-objective constraints with fewer calls. Like REINVENT 4 it is a reinforcement-learning optimiser and uses it as the comparison, but it emphasises sample efficiency.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Pretrained on a public compound library (ChEMBL 33)",
    "source_url": "https://arxiv.org/abs/2405.17066"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2405.17066"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "No wet-lab work reported",
    "source_url": "https://arxiv.org/abs/2405.17066"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Own docking multi-objective and hit-ratio tasks",
    "category": "own-task-set",
    "detail": "Docking multi-objective tasks on three receptors and hit-ratio tasks on five targets, 1,000-call oracle budget with multiple seeds; the abstract claims it beats 22 models, with baselines taken from another paper's table",
    "source_type": "self"
   },
   {
    "label": "Sample-efficiency context citation",
    "category": "comparison-vs-baseline-only",
    "detail": "PMO is cited only as context for the sample-efficiency problem; no scores on that suite are reported",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a state-space model plus memory-manipulation algorithm beats prior methods on drug-relevant multi-objective optimisation under limited oracle budgets.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation of this model on a site benchmark was found",
  "background": {
   "org_profile": "A laboratory for artificial chemical intelligence inside the chemical sciences and engineering institute of a Swiss technical university, working on AI-accelerated molecular and materials discovery and synthesis; the group was founded in 2022.",
   "funding": "Funded by two Swiss national centres of competence in research (catalysis and separations), a national science foundation starting grant and collaborations with chemical and semiconductor companies; in 2025 the group received a corporate early-career award and a university academic award.",
   "products_partnerships": "Builds on the group's earlier memory-augmented reinforcement-learning algorithm; this model is an early use of a state-space architecture in generative molecular design, released as open code, with follow-up work on synthesizability constraints.",
   "related_models": [
    "model-reinvent-4"
   ]
  },
  "reading_notes": [
   "The budget framing matters more than absolute results: the paper uses a 1,000-call oracle budget while PMO's standard budget is 10,000, so the two are not comparable.",
   "Most comparison numbers are taken from another paper's table and only some models were re-run by the authors, so versions and tuning differ.",
   "No wet lab: synthesizability and real activity are unverified.",
   "Nearly two years separate the preprint (2024-05) from the journal version (2026-03), so the comparison set may be dated."
  ],
  "caveats": [
   "Zero pairings: no scores reported on any site benchmark.",
   "The PMO pairing does not hold; the paper cites it only as background.",
   "\"Beats 22 models\" refers to the authors' own task set, not a public leaderboard position.",
   "The model name is an ordinary word, so site auto-linking must use the common-word guard."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2405.17066",
    "title": "Preprint arXiv 2405.17066",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s42256-026-01200-4",
    "title": "Journal version",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/schwallergroup/saturn",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://ai.epfl.ch/two-ai-platforms-transform-molecular-design/",
    "title": "University AI centre news",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "Candidate pairing PMO (#19)",
    "now": "Not a pairing: the original paper reports no score on this suite and cites it only as background",
    "source_url": "https://arxiv.org/abs/2405.17066",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.weights",
   "open_status.code_license",
   "release_milestones.official_announcement"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-saturn/",
   "en": "/bio/benchmark/models/saturn/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-scgpt-geneformer",
  "kind": "model",
  "short_slug": "scgpt-geneformer",
  "names": {
   "display": "scGPT / Geneformer",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "scGPT",
    "Geneformer",
    "Geneformer V2"
   ],
   "ambiguous_aliases": []
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "scGPT",
     "id_or_null": null,
     "first_release": {
      "date": "2023-05-09",
      "precision": "day",
      "event_kind": "code-release",
      "basis": "First GitHub release v0.1.2",
      "timestamp_utc": null,
      "source_url": "https://api.github.com/repos/bowang-lab/scGPT/releases",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "Geneformer",
     "id_or_null": null,
     "first_release": {
      "date": "2023",
      "precision": "year",
      "event_kind": "journal-publication",
      "basis": "Nature paper (2023); exact day not re-checked",
      "timestamp_utc": null,
      "source_url": "https://doi.org/10.1038/s41586-023-06139-9",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "Two first-generation single-cell foundation models grouped as one baseline entry; teams, corpora and versions are separate"
  },
  "family_null_reason": null,
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "First-generation single-cell foundation-model baselines for annotation, perturbation and integration",
  "org": {
   "names": [
    "University of Toronto",
    "University Health Network",
    "Vector Institute",
    "Gladstone Institutes",
    "UCSF"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CA",
   "US"
  ],
  "model_class": "Single-cell foundation model",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Both open source; Geneformer weights on Hugging Face"
  },
  "first_release": {
   "date": "2023-05",
   "precision": "month",
   "event_kind": "code-release",
   "basis": "Earliest public artifact: scGPT GitHub v0.1.2 (2023-05-09); preprint date not re-checked",
   "timestamp_utc": null,
   "source_url": "https://api.github.com/repos/bowang-lab/scGPT/releases",
   "verification": "partial",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": null,
   "journal_online": null,
   "code_release": {
    "date": "2023-05-09",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "scGPT GitHub v0.1.2",
    "timestamp_utc": null,
    "source_url": "https://api.github.com/repos/bowang-lab/scGPT/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-06-18",
    "precision": "day",
    "event_kind": "new-weights",
    "basis": "Geneformer V2 weights added on Hugging Face",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/ctheodoris/Geneformer/commits/main",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-06-18",
   "precision": "day",
   "event_kind": "new-weights",
   "basis": "Geneformer V2 models added to the Hugging Face repository",
   "description": "Geneformer V2 weights released; scGPT's last tag is v0.2.4 (2025-03-31); later commits are maintenance",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/ctheodoris/Geneformer/commits/main",
   "verification": "verified",
   "note": null
  },
  "status": "maintained",
  "status_note": "Only maintenance commits after 2025-06",
  "what_it_does": "Two first-generation single-cell foundation models: Geneformer rank-encodes genes within each cell, scGPT uses generative pretraining; both are fine-tuned for annotation, perturbation prediction and batch integration. This site uses them mainly as baselines.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Geneformer: ~29.9M human cells (later ~95M); scGPT: ~33M CELLxGENE cells",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Geneformer had one cell-level validation; scGPT had none",
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-09-scib",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "PBMC 10K",
    "source_url": "https://www.biorxiv.org/content/10.1101/2023.04.30.538439v2.full",
    "confidence": "medium",
    "verification": "partial",
    "note": "Only the scGPT paper self-reports integration; Geneformer not reported"
   },
   {
    "benchmark_id": "bench-21-dream5",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "counted": false,
    "eval_set": null,
    "source_url": "https://arxiv.org/abs/2602.17532",
    "confidence": "low",
    "verification": "unverified",
    "note": "Third-party attention analysis, abstract only; unconfirmed"
   },
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": null,
    "source_url": "https://doi.org/10.1038/s41592-025-02772-6",
    "confidence": "high",
    "verification": "verified",
    "note": "Independent studies evaluated them post hoc on public perturbation data"
   }
  ],
  "other_evaluations": [],
  "claimed_advantage": null,
  "claimed_advantage_null_reason": "Source states no developer claim",
  "independent_evidence": "Several independent 2025 studies find they do not reliably beat simple baselines on perturbation and zero-shot tasks",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "Both academic: scGPT from Toronto/UHN/Vector Institute, Geneformer from Gladstone Institutes and UCSF",
   "funding": "Geneformer funding included NIH and American Heart Association grants",
   "funding_null_reason": null,
   "products_partnerships": "Xaira Therapeutics describes scGPT as the starting point of its X-Cell model",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-x-cell",
    "model-state",
    "model-tahoe-x1",
    "model-transcriptformer",
    "model-c2s-scale",
    "model-cellfm"
   ]
  },
  "reading_notes": [
   "Appears far more often as a baseline than as a recommended method",
   "Independent evaluations show perturbation prediction no better than deliberately simple baselines"
  ],
  "caveats": [
   "Family entry: the two models were released separately; read dates per model",
   "Exact preprint and journal dates not re-checked"
  ],
  "sources": [
   {
    "url": "https://huggingface.co/ctheodoris/Geneformer/commits/main",
    "title": "Hugging Face page ctheodoris/Geneformer",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://api.github.com/repos/bowang-lab/scGPT/releases",
    "title": "GitHub repository repos/bowang-lab",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41586-023-06139-9",
    "title": "Journal article DOI 10.1038/s41586-023-06139-9",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41592-024-02201-0",
    "title": "Journal article DOI 10.1038/s41592-024-02201-0",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2023.04.30.538439v2.full",
    "title": "scGPT bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41592-025-02772-6",
    "title": "Nat Methods 2025 independent evaluation",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release",
   "family.members[1].first_release",
   "benchmarks_reported[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-scgpt-geneformer/",
   "en": "/bio/benchmark/models/scgpt-geneformer/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-scireasoner",
  "kind": "model",
  "short_slug": "scireasoner",
  "names": {
   "display": "SciReasoner",
   "full": null,
   "aliases": [],
   "ambiguous_aliases": [
    "SciReasoner"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "SciReasoner (2025 preprint)",
     "id_or_null": null,
     "first_release": {
      "date": "2025-09-25",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint v1 of an earlier paper with the same name",
      "event_kind": "paper-revision",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2509.21320",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "SciReasoner (2026 structure-reasoning version)",
     "id_or_null": "model-scireasoner",
     "first_release": {
      "date": "2026-07-08",
      "precision": "day",
      "year_hint": null,
      "basis": "Preprint v1",
      "event_kind": "paper-revision",
      "timestamp_utc": null,
      "source_url": "https://arxiv.org/abs/2607.07708",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "Two papers share the name but are different work; this record means the 2026 structure-reasoning version, and the site must state which is meant"
  },
  "family_null_reason": null,
  "domain": "text-agents",
  "domains_secondary": [
   "protein",
   "small-molecule"
  ],
  "one_liner": "Multimodal science reasoning model encoding 3D structure as shared tokens",
  "org": {
   "names": [
    "Shanghai AI Laboratory",
    "The Chinese University of Hong Kong",
    "Shanghai Institute of Materia Medica, Chinese Academy of Sciences",
    "Stanford University",
    "University of Oxford"
   ],
   "global_entity_ids": []
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "CN",
   "US",
   "GB"
  ],
  "model_class": "Multimodal science LLM (structure reasoning)",
  "open_status": {
   "code": "unclear",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The project site lists a repository but no weights or licence were visible as of 2026-11-16; evaluation code for the earlier same-named version sits in a separate public repository under a permissive licence"
  },
  "first_release": {
   "date": "2026-07-08",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1, with the repository README the same day",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2607.07708",
   "verification": "unverified",
   "note": "Two papers share the name and the intended entity must be confirmed first"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-07-08",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2607.07708",
    "verification": "unverified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-07-08",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2607.07708",
    "verification": "unverified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-07-08",
   "precision": "day",
   "year_hint": null,
   "basis": "The release event itself",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/SpectrAI-Initiative/SciReasoner",
   "verification": "unverified",
   "description": "The repository holds two commits and no releases; weights and demo are marked as coming soon and the evaluation scripts are said to arrive with the code. No later substantive event was found as of 2026-11-16.",
   "note": "Commit dates were not visible when fetched, so the recorded date may differ from the repository's by a day or two"
  },
  "status": "active",
  "status_note": "Released within 12 months with no event found since",
  "what_it_does": "Built on an open 14-billion-parameter general model, it encodes the three-dimensional structure of proteins, small molecules and inorganic crystals as shared structural tokens so that reasoning chains can cite structural evidence, covering proteins, small molecules, crystals and nucleic acids; training runs through alignment and continued pretraining, a supervised cold start and then reinforcement learning. It belongs to the same institute's science-model line as Intern-S but is a mid-size specialist rather than a general science base model, with ether0 and TxGemma as chemistry-specialist comparisons.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Dozens of public archives spanning protein sequences and structures, literature, chemistry and bioactivity, reactions, materials and nucleic acids, with proteins homologous to the test set and reactions overlapping the reaction test set removed from pretraining",
    "source_url": "https://arxiv.org/html/2607.07708"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-07-uspto",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": false,
    "note": "Self-reported exact match on that patent reaction dataset's 50,000-reaction single-step split against the previous best template-free method; the site's matrix previously recorded this pairing only as a note"
   },
   {
    "benchmark_id": "bench-17-cafa",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "medium",
    "second_hand": false,
    "note": "The main gene-ontology test set is a third-party collection rather than that competition's official blind test; the competition name appears in one figure caption with no separate score, and if kept the pairing counts as a modified protocol"
   },
   {
    "benchmark_id": "bench-14-pubmedqa",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": true,
    "note": "Absent from the paper text; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-23-bioasq",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": true,
    "note": "Absent from the paper text; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-18-labbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": true,
    "note": "Absent from the paper text; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": true,
    "note": "Absent from the paper text; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-41-bioml-bench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": true,
    "note": "Absent from the paper text; the cell rests on this site's own pages and must display as unconfirmed"
   },
   {
    "benchmark_id": "bench-42-biomysterybench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "second-hand",
    "evidence_url": "https://arxiv.org/html/2607.07708",
    "confidence": "high",
    "second_hand": true,
    "note": "Absent from the paper text; the cell rests on this site's own pages and must display as unconfirmed"
   }
  ],
  "other_evaluations": [
   {
    "label": "A large self-selected task set",
    "category": "own-task-set",
    "detail": "The paper reports 86 evaluations with best results on 67, with baselines run by the authors",
    "source_type": "self"
   },
   {
    "label": "Public tasks such as virtual screening and localisation",
    "category": "public-benchmark-not-on-site",
    "detail": "Compared with existing methods on a virtual-screening set of over a hundred targets, subcellular localisation, RNA tasks and materials properties",
    "source_type": "self"
   },
   {
    "label": "Double-blind expert assessment",
    "category": "case-study",
    "detail": "The authors report experts judging its reasoning chains better than or comparable to a frontier general model in the great majority of cases",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: encoding three-dimensional structure as shared tokens lets reasoning cite structural evidence, yielding best results on most of a large self-selected task set, with reasoning chains judged better than or comparable to a frontier general model in a double-blind expert assessment.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Weights are unreleased, no third-party evaluation was found and there is no record of official blind-test participation",
  "background": {
   "org_profile": "The lead organisation is a government-backed new-type research institution in Shanghai whose lines include a general large-model series, an evaluation suite and an open data platform; collaborators include several Chinese universities, a pharmaceutical research institute and two overseas universities.",
   "funding": "Supported by public and municipal funding with no commercial financing.",
   "products_partnerships": "The same institute runs a general science base model line recorded here as Intern-S; comparisons for this model include the chemistry specialists ether0 and TxGemma.",
   "related_models": [
    "model-intern-s",
    "model-ether0",
    "model-txgemma"
   ]
  },
  "reading_notes": [
   "The claim of best on most tasks rests on a self-selected task set with baselines the authors ran, which is not a standing on third-party boards.",
   "Gene-ontology prediction uses a third-party test collection rather than the official blind test, which requires submitting to the competition track.",
   "The comparison with general large models shows such tasks go badly without structural tokens; it does not show strength in conversational biomedical question answering.",
   "The two same-named papers must be disambiguated first, or neither first release nor latest update holds."
  ],
  "caveats": [
   "The identity between two same-named papers is unresolved and the page must state which is meant.",
   "No weights or licence were visible, so nothing can be rechecked externally.",
   "Six benchmark pairings rest on this site's own pages and must display as unconfirmed.",
   "The gene-ontology work is not official blind-test participation.",
   "The paper contains no wet-lab validation."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2607.07708",
    "title": "Preprint 2026",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2607.07708",
    "title": "Preprint full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2509.21320",
    "title": "Earlier paper with the same name, 2025",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/SpectrAI-Initiative/SciReasoner",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://scireasoner.github.io/",
    "title": "Project site",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "latest_update",
   "open_status.code",
   "open_status.weights",
   "training_data.b_private_licensed",
   "benchmarks_reported"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-scireasoner/",
   "en": "/bio/benchmark/models/scireasoner/"
  },
  "names_full_null_reason": "The paper gives no expansion; the name is the code name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1,
    "disputed-or-unconfirmed": 7
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-scprint-2",
  "kind": "model",
  "short_slug": "scprint-2",
  "names": {
   "display": "scPRINT-2",
   "full": null,
   "aliases": [
    "scPRINT-2",
    "scprint2"
   ],
   "ambiguous_aliases": [
    "scPRINT"
   ]
  },
  "family": null,
  "family_null_reason": "The predecessor scPRINT has no record and is mentioned as background",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Institut Pasteur's second-generation single-cell foundation model pretrained on 350M cells from 16 species",
  "org": {
   "names": [
    "Institut Pasteur",
    "CNRS"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "FR"
  ],
  "model_class": "Single-cell FM",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": "GPL-3.0 (GitHub) / MIT (PyPI metadata)",
   "license_changed": null,
   "note": "Code public and weights on Hugging Face; the GitHub licence is GPL-3.0 while PyPI metadata say MIT, an inconsistency"
  },
  "first_release": {
   "date": "2025-12-15",
   "precision": "day",
   "year_hint": null,
   "basis": "bioRxiv preprint v1",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.11.693702",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-12-15",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.11.693702",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-12-15",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1 (v3 on 2026-02-11)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.11.693702",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-12-17",
    "precision": "day",
    "year_hint": null,
    "basis": "First PyPI release",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://pypi.org/pypi/scprint2/json",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-09-26",
   "precision": "day",
   "year_hint": null,
   "basis": "GitHub release 2.3.8",
   "event_kind": "new-version",
   "timestamp_utc": null,
   "source_url": "https://github.com/cantinilab/scPRINT-2/releases",
   "verification": "partial",
   "description": "GitHub release 2.3.8; assets not inspected, so not to be read as new weights. PyPI is a separate version axis, latest 1.0.3 (2026-02-24)",
   "note": "Whether this release contains new weights is unverified"
  },
  "status": "active",
  "status_note": "Code release within 12 months",
  "what_it_does": "A Transformer single-cell model pretrained with masked expression prediction, denoising, contrastive cell embedding and hierarchical classification along the cell ontology; the paper first builds an \"additive\" benchmark to find which design changes help. It targets zero-shot denoising, cell embedding, cell-type prediction and gene regulatory network inference. Compared with CellFM and scGPT/Geneformer it stands out for 16-species pretraining and network inference.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Over 350M cells from more than 22k datasets and 16 species, mainly CZ CELLxGENE and scBaseCount",
    "source_url": "https://github.com/cantinilab/scPRINT-2"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://github.com/cantinilab/scPRINT-2"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-09-scib",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://www.biorxiv.org/content/10.64898/2025.12.11.693702v1",
    "confidence": "medium",
    "second_hand": false,
    "note": "The abstract claims state of the art on cell embedding but does not name scIB, and no score is on the readable pages; unconfirmed and not counted"
   },
   {
    "benchmark_id": "bench-21-dream5",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://www.biorxiv.org/content/10.64898/2025.12.11.693702v1",
    "confidence": "medium",
    "second_hand": false,
    "note": "The abstract claims better network inference and the repository mentions the benGRN package, with scores in the preprint body; protocol use unverified and not counted"
   }
  ],
  "other_evaluations": [
   {
    "label": "\"Additive\" design benchmark",
    "category": "own-task-set",
    "detail": "The authors' own benchmark used to compare design changes and finalise the model",
    "source_type": "self"
   },
   {
    "label": "Denoising and cell-type prediction",
    "category": "own-task-set",
    "detail": "Zero-shot; the abstract claims state of the art without numbers",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: state of the art on denoising, cell embedding and cell-type prediction, improved gene network inference, and generalisation to unseen modalities and species.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party leaderboard or independent evaluation result found",
  "background": {
   "org_profile": "An academic lab: the Machine Learning for Integrative Genomics junior group at Institut Pasteur, part of CNRS UMR3738, focused on single-cell data, with CNRS / ENS collaborators.",
   "funding": "The group holds an ERC Starting Grant and a completed ANR project; funding lines for this preprint are not public.",
   "products_partnerships": "The predecessor scPRINT appeared in Nature Communications in 2025-04; data loading and management use LaminDB tools; no company or pharma partnership disclosed.",
   "related_models": [
    "model-cellfm",
    "model-scgpt-geneformer",
    "model-transcriptformer",
    "model-xtrimoscperturb-scfoundation"
   ]
  },
  "reading_notes": [
   "Building a benchmark before the model is more transparent, but the readable pages have no scores, so it cannot be ranked.",
   "The state-of-the-art claim is author-reported within the authors' own benchmark scope and needs third-party confirmation.",
   "Licences differ between GitHub and PyPI; check the repository LICENSE before commercial use."
  ],
  "caveats": [
   "The scIB and DREAM5 pairings are unconfirmed and not counted.",
   "Whether GitHub 2.3.8 contains new weights is unverified.",
   "A large cell count does not imply balanced species or tissue coverage."
  ],
  "sources": [
   {
    "url": "https://www.biorxiv.org/content/10.64898/2025.12.11.693702v1",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/cantinilab/scPRINT-2",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://pypi.org/project/scprint2/",
    "title": "Package registry page",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.64898/2025.12.11.693702",
    "title": "bioRxiv version record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.nature.com/articles/s41467-025-58699-1",
    "title": "Predecessor journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://research.pasteur.fr/fr/team/machine-learning-for-integrative-genomics/",
    "title": "Research group page",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-02-24（PyPI 1.0.3）",
    "now": "2026-09-26 (GitHub release 2.3.8; PyPI is a separate version axis)",
    "source_url": "https://github.com/cantinilab/scPRINT-2/releases",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "open_status.code_license",
   "benchmarks_reported[0]",
   "benchmarks_reported[1]"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-scprint-2/",
   "en": "/bio/benchmark/models/scprint-2/"
  },
  "names_full_null_reason": "The preprint gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 2
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-seedfold",
  "kind": "model",
  "short_slug": "seedfold",
  "names": {
   "display": "SeedFold",
   "full": null,
   "full_null_reason": "No expanded name",
   "aliases": [
    "SeedFold",
    "SeedFold-Linear"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "One model from one paper; SeedFold-Linear is a variant in the same paper, not a separate entry",
  "domain": "structure",
  "domains_secondary": [],
  "one_liner": "ByteDance Seed AF3-style co-folding model that scales by widening the Pairformer, with a linear triangle-attention variant",
  "org": {
   "names": [
    "ByteDance Seed"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Co-folding (structure prediction)",
  "open_status": {
   "code": "unclear",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The paper states no code or weight licence; a public web server exists, no code or weights page found"
  },
  "first_release": {
   "date": "2025-12-30",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2512.24354 v1 submission",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2512.24354",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-12-30",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2512.24354",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-12-30",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2512.24354",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-12-30",
   "precision": "day",
   "event_kind": "none-since-release",
   "basis": "arXiv v1; project page and web search show no later event",
   "description": "No later sourced event found (as of 2026-11-16); the paper links a web server, with no code, weights page or changelog found",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2512.24354",
   "verification": "verified",
   "note": null
  },
  "status": "product-only",
  "status_note": "Paper and public web server only; no code or weights",
  "what_it_does": "Keeps the AlphaFold 3 Pairformer-plus-diffusion design but argues performance is limited by hidden width, so it widens the Pairformer; SeedFold-Linear swaps triangle attention for a linear version to cut cost. Where the same company's Protenix reproduces AF3 through data and recipes, SeedFold contributes architecture scaling; SeedProteo later reuses its folding architecture for design.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "PDB 180,540, AFDB 3,326,991 and MGnify distillation 23,075,211 entries, cutoff 2021-09-30",
    "source_url": "https://arxiv.org/abs/2512.24354"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-31-foldbench",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "counted": true,
    "eval_set": "FoldBench categories (protein-ligand, antibody-antigen, protein-protein, RNA)",
    "source_url": "https://arxiv.org/abs/2512.24354",
    "confidence": "high",
    "verification": "verified",
    "note": "Self-reported in the paper, not run by the board"
   }
  ],
  "other_evaluations": [
   {
    "label": "Width and linear-attention ablations",
    "category": "own-task-set",
    "detail": "In-paper comparison of the two variants and scaling choices",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: widening the Pairformer beats stacking layers, and linear triangle attention cuts compute while keeping accuracy.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party replication or board-run result found",
  "background": {
   "org_profile": "ByteDance Seed is ByteDance's foundation-model research division; its AI for Science work covers biological foundation models, quantum chemistry and molecular dynamics. A second-hand report (2026-06) says an AI drug-discovery unit absorbing the protein-structure team is being spun out under ByteDance control; whether SeedFold moves is not stated.",
   "funding": "Funded from ByteDance's internal budget; no separate fundraising.",
   "funding_null_reason": null,
   "products_partnerships": "The same company open-sources the Protenix series and PXDesign; SeedProteo reuses the SeedFold architecture. No pharma partnerships disclosed.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-protenix",
    "model-seedproteo",
    "model-alphafold3",
    "model-boltz-2"
   ]
  },
  "reading_notes": [
   "The two variants trade wins across complex categories; on RNA the paper's own table still favours AF3.",
   "All numbers are self-reported on FoldBench without third-party replication and are not comparable with PoseBusters-type figures.",
   "Only public data is used, so gains come from architecture and scale rather than private data."
  ],
  "caveats": [
   "Weight availability is not stated; practical access is unverified.",
   "Same company as Protenix but a different team; their relationship is not described officially.",
   "Organizational spin-out information comes from second-hand reports only."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2512.24354",
    "title": "SeedFold arXiv 2512.24354",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://seedfold.github.io",
    "title": "SeedFold project page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2512.24192",
    "title": "SeedProteo arXiv 2512.24192",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://seed.bytedance.com/en/direction/ai_for_science",
    "title": "Seed AI for Science page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://kr-asia.com/bytedance-spins-out-drug-discovery-unit-to-test-ai4s-commercialization",
    "title": "Report on ByteDance spinning out its drug-discovery unit (second-hand)",
    "kind": "second-hand",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "benchmarks_reported",
    "was": "No direct site benchmark pairing; FoldBench outside scope",
    "now": "FoldBench is now a site entry; counted as self-reported under the official protocol",
    "source_url": "https://arxiv.org/abs/2512.24354",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.code",
   "open_status.weights"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-seedfold/",
   "en": "/bio/benchmark/models/seedfold/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-seedproteo",
  "kind": "model",
  "short_slug": "seedproteo",
  "names": {
   "display": "SeedProteo",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "SeedProteo"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "protein",
  "domains_secondary": [],
  "one_liner": "Diffusion-based all-atom de novo protein design model for binders",
  "org": {
   "names": [
    "ByteDance Seed"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "De novo binder design",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Web server only; no weights or code listed"
  },
  "first_release": {
   "date": "2025-12-30",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2512.24192 v1 submitted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2512.24192",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-12-30",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2512.24192",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-02-24",
   "precision": "day",
   "event_kind": "new-version",
   "basis": "arXiv v2 adding in vitro results",
   "description": "arXiv v2 adds wet-lab results; no code or weights found, access only via the web server",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2512.24192",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "New paper version within 12 months",
  "what_it_does": "Adapts the SeedFold folding architecture into a generator that co-designs sequence and structure for de novo target binders. Available only through a web server.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "~500k monomers from public structure databases, ~50k two-chain PDB complexes, plus ~100k synthetic domain-interaction pairs",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Wet lab used only for validation on two targets",
    "source_url": null
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Binding assays on two targets",
    "modality": "protein binder",
    "targets": 2,
    "designs_tested": 20,
    "hits": null,
    "denominator_basis": "10 designs each on two targets, all expressed; the abstract gives a 70–80% hit range, while per-target hit counts and exact best affinities sit in figures and are not cited",
    "success_rate": "abstract states a 70-80% hit rate",
    "affinity_best": "picomolar for PD-L1 and sub-nanomolar for RBD (range only)",
    "version": "arXiv v2",
    "source_type": "CRO",
    "source_url": "https://arxiv.org/html/2512.24192v2",
    "date": null,
    "verification": "verified",
    "note": "The paper states a CRO executed the assays; not an independent replication. Only ranges are given, not the four figure values"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Unconditional monomer generation",
    "category": "own-task-set",
    "detail": "Lengths 100–1000 and three topologies, compared with several open generative models",
    "source_type": "self"
   },
   {
    "label": "In silico binder design",
    "category": "own-task-set",
    "detail": "10 targets with about 1,000 designs each, judged by its own folding model",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report the highest in silico binder design success among open methods and a 70–80% wet-lab hit range on two targets",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Third-party work does not name it",
  "background": {
   "org_profile": "The developer is ByteDance's foundation-model research division, whose AI for Science work also includes the open-source Protenix; internally funded",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "Wet-lab assays were bought from a CRO; press reports say its AI drug discovery unit is being spun out, without naming this model",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-seedfold",
    "model-protenix",
    "model-boltzgen",
    "model-la-proteina"
   ]
  },
  "reading_notes": [
   "The wet lab was CRO-executed and the discriminator is the same team's folding model; not independent validation",
   "Only two targets with 10 designs each: small scale, and per-target values cannot be cited"
  ],
  "caveats": [
   "Closed, web-service only; cannot be reproduced externally",
   "The abstract's hit range and the figure values differ; the range is recorded"
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2512.24192",
    "title": "arXiv preprint 2512.24192",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://seedfold.github.io/",
    "title": "Official page seedfold.github.io",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-seedproteo/",
   "en": "/bio/benchmark/models/seedproteo/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-stack",
  "kind": "model",
  "short_slug": "stack",
  "names": {
   "display": "Stack",
   "full": null,
   "aliases": [
    "Stack-Large"
   ],
   "ambiguous_aliases": [
    "Stack"
   ]
  },
  "family": null,
  "family_null_reason": "Only one public size exists, and the companion prediction set is not a separate model",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Single-cell foundation model doing in-context learning from example cells in the prompt, without fine-tuning",
  "org": {
   "names": [
    "Arc Institute"
   ],
   "global_entity_ids": [
    "arc-institute"
   ]
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Single-cell foundation model (in-context learning)",
  "open_status": {
   "code": "open",
   "weights": "non-commercial",
   "weights_license": "non-commercial licence",
   "code_license": null,
   "license_changed": null,
   "note": "The largest size's weights are on the model hub under a non-commercial licence; code is public and the licence text was not retrieved this round"
  },
  "first_release": {
   "date": "2026-01-09",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v1 posted (10.64898/2026.01.09.698608)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.64898/2026.01.09.698608",
   "verification": "verified",
   "note": "Code v0.1.1 followed the next day (2026-01-10)"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-01-09",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.64898/2026.01.09.698608",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-01-09",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.64898/2026.01.09.698608",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-01-10",
    "precision": "day",
    "year_hint": null,
    "basis": "GitHub v0.1.1",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/ArcInstitute/stack/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2026-01-09",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights on the model hub at launch under a non-commercial licence",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://arcinstitute.org/news/foundation-model-stack",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-01-09",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer launch page",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://arcinstitute.org/news/foundation-model-stack",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-06-08",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint v2 posted",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.64898/2026.01.09.698608",
   "verification": "verified",
   "description": "Preprint v2 posted; the earlier event is GitHub v0.1.3 (2026-03-16, dependency fixes), the last release. No journal version found",
   "note": "The latest update was recorded as maintenance-only; corrected to the later paper revision"
  },
  "status": "active",
  "status_note": "Paper revision and a version release within 12 months",
  "what_it_does": "After pretraining, reference cells in the prompt express a condition (a perturbation, a disease state) that the model transfers to target cells without updating weights; a companion release contains about twenty thousand predicted responses across cell type, tissue and perturbation combinations, with tissue lineages taken from a public human atlas. Unlike the institute's STATE, it is a general in-context model rather than a response predictor trained on perturbation data.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Pretrained on 149 million cells from a public count collection; post-trained on 55 million cells from a public cell atlas and peripheral-blood perturbation data produced by a reagent company",
    "source_url": "https://arcinstitute.org/news/foundation-model-stack"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "None by the developer; the perturbation data used in post-training was produced by a third-party company",
    "source_url": "https://arcinstitute.org/news/foundation-model-stack"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The companion prediction set has no wet-lab validation",
    "source_url": "https://arcinstitute.org/news/foundation-model-stack"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.64898/2026.01.09.698608",
    "confidence": "high",
    "second_hand": false,
    "note": "Developer-defined perturbation sets against two classical methods trained from scratch per set plus several foundation models; neither the launch page nor the preprint gives citable numbers. The challenge claim does not appear on the launch page"
   }
  ],
  "other_evaluations": [
   {
    "label": "Large predicted-response set",
    "category": "case-study",
    "detail": "About twenty thousand predicted combinations, all prediction with no wet-lab validation: a hypothesis library, not facts",
    "source_type": "self"
   },
   {
    "label": "Conflicting perturbation counts",
    "category": "own-task-set",
    "detail": "Different versions of the material give two perturbation totals while the launch page gives only combination counts; cite the source version",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: in-context learning transfers conditions without fine-tuning and consistently beats classical methods trained from scratch and other foundation models on perturbation prediction.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation or replication was found, and the independent literature's findings about linear and mean baselines on this task have not been tested on this model",
  "background": {
   "org_profile": "A California non-profit research institute founded in 2021 with over $650M committed at launch; core investigators hold multi-year renewable terms alongside university appointments. No citable official headcount, so none is given.",
   "funding": "Non-profit and endowment-funded; 2024 budget figures differ between sources, so no single number is given.",
   "products_partnerships": "The pretraining dataset is institute-built as part of its cell-atlas initiative; the post-training perturbation data comes from a third-party reagent company. Sibling models are STATE and Evo 2 under the same virtual-cell initiative.",
   "related_models": [
    "model-state",
    "model-evo-2",
    "model-tahoe-x1"
   ]
  },
  "reading_notes": [
   "Comparisons are mainly against two classical methods and other foundation models; the independent literature finds linear and mean baselines hard to beat on perturbation prediction, and those baselines are absent here.",
   "The roughly twenty thousand predicted responses have no wet-lab validation and are not established biology.",
   "No public score exists on the challenge, so there is no standing to speak of.",
   "The perturbation data came from a third-party company and must not be attributed to the developer."
  ],
  "caveats": [
   "The only site pairing is self-reported under a modified protocol with no citable numbers.",
   "No journal version; only two preprint versions.",
   "Perturbation totals conflict between materials.",
   "Weights carry a non-commercial licence."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.64898/2026.01.09.698608",
    "title": "Preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.64898/2026.01.09.698608",
    "title": "Preprint version record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arcinstitute.org/news/foundation-model-stack",
    "title": "Developer launch page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/ArcInstitute/stack/releases",
    "title": "Code repository releases",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-03-16, code maintenance only",
    "now": "2026-06-08, preprint v2 (paper revision)",
    "source_url": "https://api.crossref.org/works/10.64898/2026.01.09.698608",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Claims related to the challenge",
    "now": "The release page does not mention the challenge and gives no score; only the modified-protocol self-report on the developers' own perturbation split is kept",
    "source_url": "https://arcinstitute.org/news/foundation-model-stack",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.weights_release",
   "open_status.code_license",
   "other_evaluations[1]"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-stack/",
   "en": "/bio/benchmark/models/stack/"
  },
  "names_full_null_reason": "The preprint gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developing institute but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-state",
  "kind": "model",
  "short_slug": "state",
  "names": {
   "display": "STATE",
   "full": "State Transition + State Embedding",
   "aliases": [
    "STATE model"
   ],
   "ambiguous_aliases": [
    "STATE",
    "State"
   ]
  },
  "family": null,
  "family_null_reason": "The developers say a next generation is in development but unreleased; no family entry",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Virtual cell model predicting transcriptional response to drug or genetic perturbation across cell lines",
  "org": {
   "names": [
    "Arc Institute"
   ],
   "global_entity_ids": [
    "arc-institute"
   ]
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Single-cell perturbation prediction / virtual cell",
  "open_status": {
   "code": "open",
   "weights": "non-commercial",
   "weights_license": "non-commercial licence",
   "code_license": null,
   "license_changed": null,
   "note": "Code is public and weights carry a non-commercial licence; the full licence text was not retrieved this round"
  },
  "first_release": {
   "date": "2025-06-26",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint (10.1101/2025.06.26.661135) and the developer's launch page",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arcinstitute.org/news/virtual-cell-model-state",
   "verification": "partial",
   "note": "The site also records a 2025-06-27 posting; the day-level difference is unresolved"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-06-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint and launch page in the same window",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arcinstitute.org/news/virtual-cell-model-state",
    "verification": "partial",
    "note": null
   },
   "paper_v1": {
    "date": "2025-06-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1101/2025.06.26.661135",
    "verification": "partial",
    "note": null
   },
   "journal_online": {
    "date": "2026-09",
    "precision": "month",
    "year_hint": null,
    "basis": "Journal issue (registration is month-level; no online-first day found)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1016/j.cell.2026.07.052",
    "verification": "partial",
    "note": "Not padded to a day"
   },
   "code_release": {
    "date": "2026-07",
    "precision": "month",
    "year_hint": null,
    "basis": "Repository v0.11.x releases (2026-06 to 07)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/ArcInstitute/state/releases",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-06-26",
    "precision": "day",
    "year_hint": null,
    "basis": "Developer launch page",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://arcinstitute.org/news/virtual-cell-model-state",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-09",
   "precision": "month",
   "year_hint": null,
   "basis": "Journal publication (issue is month-level)",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1016/j.cell.2026.07.052",
   "verification": "partial",
   "description": "Journal version published; the repository's v0.11.x releases (2026-06 to 07) are code releases. The developers say a next generation is in development but unreleased",
   "note": "No online-first day was found; only month-level evidence"
  },
  "status": "active",
  "status_note": "Journal publication and code releases within 12 months",
  "what_it_does": "Learns state transitions between pre- and post-perturbation cells to predict transcriptional response to drugs or genetic perturbation across cell lines, with a developer-built evaluation tool alongside. Unlike the institute's Stack, which does prompt-style prediction by in-context learning, this model trains on perturbation data for response prediction; it shares the same third-party perturbation corpus as Tahoe-x1.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 167 million observational cells (including the institute's atlas and a public count collection), over 100 million perturbed cells across 70 cell lines, plus a public Perturb-seq dataset",
    "source_url": "https://arcinstitute.org/news/virtual-cell-model-state"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "None for the developer: the main perturbation corpus was produced by a data company and donated to the institute's atlas (about 100 million cells, 50 cancer cell lines, 1,100 perturbations) and another set came from a reagent company; the institute's own challenge data is used only for evaluation. To avoid leakage, training used 20 of the cell lines with 5 held out",
    "source_url": "https://arcinstitute.org/news/arc-vevo"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1016/j.cell.2026.07.052"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Perturbation data description (not an own experiment)",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "The same third-party perturbation corpus; the perturbation count is unified to 1,100 this round (the press release's ~1,200 is not used)",
    "success_rate": null,
    "affinity_best": null,
    "version": "primary paper",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398",
    "date": {
     "date": "2025-06-26",
     "precision": "day",
     "year_hint": null,
     "basis": "Data description alongside this model's preprint",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Data produced by a third-party company, not a wet-lab campaign by the developer",
    "unverified_marker": null,
    "note": "This item exists only to correct the published perturbation count and is not wet-lab validation of the model"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://arcinstitute.org/news/virtual-cell-model-state",
    "confidence": "high",
    "second_hand": false,
    "note": "Evaluated on the third-party perturbation corpus with the developer's own metric; no blind challenge score is published and nothing public shows the blind protocol was followed, so that part stays unconfirmed"
   }
  ],
  "other_evaluations": [
   {
    "label": "Developer-built evaluation tool",
    "category": "own-task-set",
    "detail": "The official page and the paper abstract frame the same result two ways (about 50% and over 30%) and claim it is the first model to beat linear baselines consistently",
    "source_type": "self"
   },
   {
    "label": "Linear-baseline findings in independent evaluations",
    "category": "comparison-vs-baseline-only",
    "detail": "Several independent evaluations and a challenge wrap-up find simple baselines hard to beat, cutting against the developer's claim",
    "source_type": "third"
   },
   {
    "label": "scIB",
    "category": "public-benchmark-not-on-site",
    "detail": "Confirmed not a pairing: neither the paper nor the official page reports that benchmark",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: learning state transitions substantially improves separation of perturbation effects and recovery of truly differential genes, and beats linear baselines consistently for the first time.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "No independent replication of this model was found; related independent evaluations and the perturbation-prediction challenge wrap-up conclude that simple baselines are hard to beat.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "A California non-profit research institute founded in 2021 with over $650M committed at launch; core investigators hold multi-year renewable terms alongside university appointments. No citable official headcount, so none is given.",
   "funding": "Non-profit and endowment-funded; 2024 budget figures differ between sources, so no single number is given.",
   "products_partnerships": "The institute's virtual-cell initiative includes a cell atlas and an annual perturbation-prediction challenge (first launched 2025-06-26 with a $100k grand prize and three sequencing and compute sponsors; a 2026 edition is announced); sibling models are Stack and Evo 2; the main perturbation corpus was donated by a data company that also builds Tahoe-x1.",
   "related_models": [
    "model-stack",
    "model-tahoe-x1",
    "model-evo-2"
   ]
  },
  "reading_notes": [
   "The same result is framed two ways on the official page and in the abstract; always say which framing is cited.",
   "Training and evaluation share one perturbation corpus; even with held-out cell lines this stays a within-distribution comparison.",
   "The main wet-lab data came from a third-party company and must not be attributed to the developer.",
   "No blind score for this model is published for the challenge, so its blind-test standing is unknown."
  ],
  "caveats": [
   "The only site pairing is self-reported under a modified protocol, and the blind-challenge part is unconfirmed.",
   "The journal online date has month-level evidence only.",
   "The perturbation count conflicts between sources; the lower framing is adopted this round.",
   "Weights carry a non-commercial licence."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1016/j.cell.2026.07.052",
    "title": "Journal paper 2026",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2025.06.26.661135",
    "title": "Preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arcinstitute.org/news/virtual-cell-model-state",
    "title": "Developer launch page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://github.com/ArcInstitute/state/releases",
    "title": "Code repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up",
    "title": "Challenge wrap-up",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://arcinstitute.org/news/arc-vevo",
    "title": "Perturbation data donation announcement",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "training_data.c_own_wetlab_for_training",
    "was": "Perturbation data on about 1,200 drugs",
    "now": "1,100 perturbations (about 1,200 appears only in press releases)",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "Challenge results (treated as a board)",
    "now": "Modified-protocol self-report; the challenge's blind-test scores are not published, marked unconfirmed",
    "source_url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "first_release",
   "latest_update",
   "release_milestones.journal_online",
   "benchmarks_reported[0]",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-state/",
   "en": "/bio/benchmark/models/state/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developing institute but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-tahoe-x1",
  "kind": "model",
  "short_slug": "tahoe-x1",
  "names": {
   "display": "Tahoe-x1",
   "full": null,
   "aliases": [
    "Tahoe-x1",
    "Tx1-70M",
    "Tx1-1.3B",
    "Tx1-3B"
   ],
   "ambiguous_aliases": [
    "Tahoe"
   ]
  },
  "family": null,
  "family_null_reason": "The three parameter sizes are configurations of one model, not family members",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Single-cell foundation model: masked-pretraining encoder trained at scale including perturbation data",
  "org": {
   "names": [
    "Tahoe Therapeutics"
   ],
   "global_entity_ids": [
    "tahoe-therapeutics"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Single-cell foundation model (masked-pretraining encoder)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "All three weight sizes and the code are public under a permissive licence"
  },
  "first_release": {
   "date": "2025-10-10",
   "precision": "day",
   "year_hint": null,
   "basis": "First code and weights release (v0.1.0); the preprint followed on 2025-10-23",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/tahoebio/tahoe-x1/releases",
   "verification": "verified",
   "note": "The model-hub repository was created 2025-08-25 with a README only; repository creation is not used as first release"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-10-10",
    "precision": "day",
    "year_hint": null,
    "basis": "v0.1.0 release",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/tahoebio/tahoe-x1/releases",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-10-23",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1 (10.1101/2025.10.23.683759)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2025.10.23.683759",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-10-10",
    "precision": "day",
    "year_hint": null,
    "basis": "GitHub v0.1.0",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/tahoebio/tahoe-x1/releases",
    "verification": "verified",
    "note": null
   },
   "weights_release": {
    "date": "2025-10-10",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights released with the code under a permissive licence",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/tahoebio/Tahoe-x1",
    "verification": "partial",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-06-03",
   "precision": "day",
   "year_hint": null,
   "basis": "Perturbation tutorial added to the repository",
   "event_kind": "code-maintenance-only",
   "timestamp_utc": null,
   "source_url": "https://github.com/tahoebio/tahoe-x1/releases",
   "verification": "verified",
   "description": "A perturbation tutorial was added to the repository, a maintenance change; no new version or journal publication was found this round. This is the latest sourced event",
   "note": null
  },
  "status": "maintained",
  "status_note": "Maintenance commits only, with no new version or journal publication",
  "what_it_does": "Masked language modelling with an encoder, with perturbation data deliberately included in pretraining; downstream tasks span gene essentiality, cell-type classification and perturbation response. The developers claim several-fold to tens-of-fold better compute efficiency than existing cell-state models - a company statement, not independently verified. It shares a perturbation corpus with STATE but is a representation-learning foundation model rather than a response predictor, and differs again from the in-context route of Stack.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Two public single-cell atlases and count collections, totalling 266 million single-cell profiles with the in-house corpus",
    "source_url": "https://huggingface.co/tahoebio/Tahoe-x1"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "In-house perturbation corpus: about 100 million profiles, 1,100 small-molecule perturbations, 50 cancer cell lines; the data producer and the model developer are the same company",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1101/2025.10.23.683759"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "In-house perturbation corpus scale (training data, not a validation campaign)",
    "targets": null,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "About 100 million profiles, 1,100 small-molecule perturbations, 50 cancer cell lines; 266 million cells in pretraining",
    "success_rate": null,
    "affinity_best": null,
    "version": "primary paper v3",
    "source_type": "self",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398",
    "date": {
     "date": "2025-10-23",
     "precision": "day",
     "year_hint": null,
     "basis": "Data description alongside this model's preprint",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "https://api.crossref.org/works/10.1101/2025.10.23.683759",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Produced on the developer's own experimental platform, used for training rather than validation",
    "unverified_marker": null,
    "note": "The perturbation count is 1,100 this round; the ~1,200 figure circulating in press releases is not used, and the same fix applies to the training-data line of STATE"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://doi.org/10.1101/2025.10.23.683759",
    "confidence": "high",
    "second_hand": false,
    "note": "Held-out-context perturbation response under a developer-defined setup explicitly stated not to be the challenge; neither the abstract nor the model card gives numbers"
   }
  ],
  "other_evaluations": [
   {
    "label": "Public gene-essentiality dataset",
    "category": "public-benchmark-not-on-site",
    "detail": "Global and context-specific essentiality prediction on public CRISPR screening data, which is a different task from perturbation response; the abstract and model card state conclusions without numbers",
    "source_type": "self"
   },
   {
    "label": "Cancer hallmark genes and cell-type classification",
    "category": "own-task-set",
    "detail": "The abstract claims state-of-the-art on all four tasks with no citable figures",
    "source_type": "self"
   },
   {
    "label": "Compute-efficiency claim",
    "category": "comparison-vs-baseline-only",
    "detail": "The company claims several-fold to tens-of-fold higher efficiency, not independently verified",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: including their own perturbation data in pretraining puts the model at state of the art on essentiality, hallmark genes, cell typing and held-out-context perturbation response at once, with better compute efficiency.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation or replication was found; another institute's model uses the same data producer's corpus, but that is not an evaluation of this model",
  "background": {
   "org_profile": "A single-cell data and model company founded in 2022 on technology from a US university, headquartered in California; sources date its renaming to either 2024-04 or 2025-04. Its core assets are its experimental platform and in-house perturbation dataset. No citable headcount, so none is given.",
   "funding": "A $12M seed in 2022-12 and a $30M Series A in 2025-08 (one fund leading with six further investors), $42M in total, with the round earmarked for a billion-scale single-cell dataset.",
   "products_partnerships": "Sample prep and sequencing for the in-house corpus were done by two instrument and reagent companies, and the corpus was open-sourced as the first dataset in a research institute's cell atlas; in 2026-01 the company, that institute and a biohub announced a larger joint perturbation dataset (press-release framing: over 120 million single-cell data points, about 50 cell lines, 1,400 chemical scaffolds at three doses each and 100 cytokine perturbations, shared among the three parties before open release); also in 2026-01 it announced a joint venture with an antibody platform company for antibody-drug conjugates, starting with two programmes and undisclosed terms. On-site relation: STATE trains on the same corpus.",
   "related_models": [
    "model-state",
    "model-stack",
    "model-x-cell"
   ]
  },
  "reading_notes": [
   "The essentiality evaluation uses public CRISPR data, a different task from perturbation response, and the four state-of-the-art claims come without numbers.",
   "Producing its own data removes third-party dependence but brings the risk that training and evaluation share one distribution.",
   "The efficiency multiple is a company statement that has not been independently verified.",
   "With no blind challenge score it cannot be compared with other perturbation models on the same blind test."
  ],
  "caveats": [
   "The only site pairing is self-reported under a modified protocol that the developers say is not the challenge.",
   "No journal version; the latest event is repository maintenance only.",
   "Perturbation counts conflict between press releases and the paper; the paper's figure is used.",
   "In-house data enters the training set, so the within-distribution risk must be stated when reading results."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1101/2025.10.23.683759",
    "title": "Preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://huggingface.co/tahoebio/Tahoe-x1",
    "title": "Weights and model card",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://github.com/tahoebio/tahoe-x1/releases",
    "title": "Code repository releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398",
    "title": "In-house perturbation corpus paper",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arcinstitute.org/news/arc-vevo",
    "title": "Corpus open-sourcing announcement (third-party institute)",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "training_data.c_own_wetlab_for_training",
    "was": "About 1,100–1,200 drugs",
    "now": "1,100 small-molecule perturbations (about 1,200 appears only in press releases)",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.02.20.639398",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2025-10 (preprint)",
    "now": "2025-10-10 (first code and weights version)",
    "source_url": "https://github.com/tahoebio/tahoe-x1/releases",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.weights_release",
   "other_evaluations[2]",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-tahoe-x1/",
   "en": "/bio/benchmark/models/tahoe-x1/"
  },
  "names_full_null_reason": "The preprint and model card use the product name only",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developing company but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-tamgen",
  "kind": "model",
  "short_slug": "tamgen",
  "names": {
   "display": "TamGen",
   "full": null,
   "aliases": [
    "TamGen"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single model with no family members or successor",
  "domain": "small-molecule",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Target-aware molecule generator that conditions on a protein pocket and refines compounds",
  "org": {
   "names": [
    "Microsoft Research AI for Science",
    "Global Health Drug Discovery Institute (GHDDI)"
   ],
   "global_entity_ids": []
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "CN",
   "US"
  ],
  "model_class": "Target-conditioned molecule generation (chemical language model + pocket encoder)",
  "open_status": {
   "code": "open",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code is open under the developer's GitHub organisation; licence text and weights release were not retrieved this round"
  },
  "first_release": {
   "date": "2024-01-08",
   "precision": "day",
   "year_hint": null,
   "basis": "Preprint posted (registry record)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1101/2024.01.08.574635",
   "verification": "verified",
   "note": "Previously recorded as 2024-01 (month only); upgraded to day precision from the registry record"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-01-08",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2024.01.08.574635",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-01-08",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2024.01.08.574635",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2024-10-29",
    "precision": "day",
    "year_hint": null,
    "basis": "Journal publication (DOI 10.1038/s41467-024-53632-4)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41467-024-53632-4",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2024-10-29",
   "precision": "day",
   "year_hint": null,
   "basis": "Journal publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41467-024-53632-4",
   "verification": "partial",
   "description": "Journal publication is the latest sourced substantive event; the repository has no releases and its last commit (2025-09-24) is maintenance whose content was not inspected",
   "note": "Maintenance commits post-date the journal event, but the policy on counting maintenance commits is unsettled, so this is flagged unverified"
  },
  "status": "maintained",
  "status_note": "Maintenance commits only, the last on 2025-09-24",
  "what_it_does": "Adds a protein-pocket context encoder to a chemical language model for target-conditioned generation and refinement of existing molecules; the paper runs a full design-refine-test loop, with the big-lab side generating compounds and a nonprofit drug-research institute doing binding analysis, structure-activity work and enzyme assays. Unlike GenMol and Saturn it closes a real wet-lab loop but does not compete on public optimisation suites.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Decoder pretrained on 10 million SMILES sampled from a public compound library; target conditioning trained on CrossDocked2020 and similar complex sets",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "No claim of feeding assay results back into training",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "In-vitro enzyme inhibition against one tuberculosis protease target, used for validation only",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "In-vitro protease inhibition, tuberculosis target",
    "targets": 1,
    "designs_tested": 296,
    "hits": 14,
    "denominator_basis": "2,612 / 8,635 designs generated with 296 selected onward; the 14 active compounds comprise 5 commercial analogues, 3 structure-activity syntheses, and 6 of 8 directly synthesised compounds (in-vitro inhibition below the 40 μM level)",
    "success_rate": null,
    "affinity_best": "about 1.9 μM, but that compound is a commercial analogue, not a model-generated molecule",
    "version": "journal paper",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/",
    "date": {
     "date": "2024-10-29",
     "precision": "day",
     "year_hint": null,
     "basis": "Journal version",
     "event_kind": "journal-publication",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1038/s41467-024-53632-4",
     "verification": "verified",
     "note": null
    },
    "verification": "partial",
    "executor": "Assays run by the partner nonprofit drug-research institute; not an independent evaluation",
    "unverified_marker": null,
    "note": "No overall success rate: the hits span three different sources (commercial analogues, SAR syntheses, direct syntheses) and cannot be merged into one ratio; the published \"16 tested\" claim does not exist in the paper and is removed"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-32-crossdocked",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/",
    "confidence": "medium",
    "second_hand": false,
    "note": "100 test pockets compared with several 3D generative methods on docking and drug-likeness metrics; per-metric values appear only in a figure and the supplement, not the main text"
   }
  ],
  "other_evaluations": [
   {
    "label": "Own metric combination for target-conditioned generation",
    "category": "own-task-set",
    "detail": "The paper aggregates drug-likeness and docking metrics with a mean reciprocal rank; that aggregation is not an official protocol",
    "source_type": "self"
   },
   {
    "label": "Viral target case",
    "category": "case-study",
    "detail": "The developer page also mentions a coronavirus target; a case description without evaluation data",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: target-conditioned generation plus refinement yields compounds that can go straight to assay, and ranks near the top on several drug-likeness and docking metrics on a public pocket set.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation or replication found; assays were run by the partner institute and compiled by the authors",
  "background": {
   "org_profile": "A science-AI research division of a US software company (founded 2022) together with a nonprofit drug-discovery institute founded in Beijing in 2016 by a global-health foundation, the Beijing government and a Chinese university; co-authors also come from two Chinese universities.",
   "funding": "The nonprofit institute received about $100M of five-year support initially, and in 2023-06 the foundation announced a further $50M over five years with municipal matching; the paper states partial support from that foundation.",
   "products_partnerships": "The two organisations announced a collaboration in 2024-01: the big-tech lab generates candidates and the institute runs the experiments, with stated plans for lead optimisation, ADMET improvement and preclinical work; terms undisclosed.",
   "related_models": [
    "model-genmol",
    "model-saturn"
   ]
  },
  "reading_notes": [
   "The 14 actives are not 14 model-generated molecules: five are commercial analogues and three are structure-activity syntheses, and the most potent one is a commercial analogue.",
   "The real denominators are 2,612 / 8,635 designs with 296 selected; the site's earlier \"16 tested\" is not in the paper.",
   "In-vitro enzyme inhibition is early hit-level evidence; no cell or animal data, so it is far from a drug.",
   "On CrossDocked the per-metric values live in a figure and the supplement with none in the main text, so cross-paper comparison needs care."
  ],
  "caveats": [
   "Wet-lab hits span three source types and cannot be merged into a single success rate.",
   "The latest-update framing depends on the unsettled maintenance-commit policy and is flagged unverified.",
   "The site said there was no site-benchmark pairing; one self-reported pairing on the official protocol does exist.",
   "The aggregate ranking method is the authors' own, not the benchmark's official protocol."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1038/s41467-024-53632-4",
    "title": "Journal paper 2024",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/",
    "title": "Journal full text (public archive)",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1101/2024.01.08.574635",
    "title": "Preprint registry record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/microsoft/TamGen",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://www.microsoft.com/en-us/research/?p=1105683",
    "title": "Developer research blog",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "16 compounds tested, 14 strong inhibitors, best IC50 1.88 μM",
    "now": "Removed \"16 tested\" (the paper does not say so); the 14 actives are 5 commercial analogues + 3 SAR syntheses + 6 of 8 direct syntheses, and the most potent is a commercial analogue; 2,612/8,635 designed, 296 selected",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2024-01",
    "now": "2024-01-08 (preprint posted)",
    "source_url": "https://api.crossref.org/works/10.1101/2024.01.08.574635",
    "date_found": "2026-11-16"
   },
   {
    "field": "benchmarks_reported",
    "was": "No pairing with a benchmark on this site",
    "now": "CrossDocked: self-reported on the official protocol",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11522292/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "open_status.weights",
   "open_status.code_license",
   "wetlab_validation[0].affinity_best"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-tamgen/",
   "en": "/bio/benchmark/models/tamgen/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-terrabind",
  "kind": "model",
  "short_slug": "terrabind",
  "names": {
   "display": "TerraBind",
   "full": null,
   "aliases": [
    "TerraBind",
    "EMMI Predict"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Only one version (arXiv v1); no family members",
  "domain": "small-molecule",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Protein-ligand structure and affinity foundation model trading diffusion for speed",
  "org": {
   "names": [
    "Terray Therapeutics"
   ],
   "global_entity_ids": []
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Co-folding / affinity prediction (diffusion-free)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "The paper gives no code or weights link, so it is treated as not open; it is offered under a commercial product name"
  },
  "first_release": {
   "date": "2026-02-08",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv v1 submitted (the only version)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2602.07735",
   "verification": "verified",
   "note": "The company press release came four days later and is not used as first-release evidence"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-02-08",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2602.07735",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-02-08",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2602.07735",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-02-12",
    "precision": "day",
    "year_hint": null,
    "basis": "Commercial product launch press release (a promotional event, not a new version)",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.businesswire.com/news/home/20260212470275/en",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-02-12",
   "precision": "day",
   "year_hint": null,
   "basis": "Commercial product launch press release",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://www.businesswire.com/news/home/20260212470275/en",
   "verification": "partial",
   "description": "The latest sourced event is a press release for the commercial product launch; no later version, code or weights were found and the arXiv record is still v1. The event is promotional, not a new model version",
   "note": "Must display as unverified: a press release is not a model version event"
  },
  "status": "product-only",
  "status_note": "No public code or weights; offered as a commercial product",
  "what_it_does": "Uses a smaller pairformer trunk with a diffusion-free optimisation scheme to output structure and affinity directly, plus a per-inference uncertainty estimate; parameter comparisons hold only at the trunk level and must not be set against the baseline's full model size. Nesso-1 explicitly follows this slimming route, while AQAffinity builds on another open co-folding model; the comparison baseline is Boltz-2.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public structure and predicted-structure databases, public complex sets, with affinity from several public activity databases plus around a million synthetic decoys",
    "source_url": "https://arxiv.org/html/2602.07735"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "None for the base model; nine proprietary crystal structures across two targets were used only in a separate fine-tuning experiment",
    "source_url": "https://arxiv.org/html/2602.07735"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "The paper states the base model uses public data only",
    "source_url": "https://arxiv.org/html/2602.07735"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Existing biochemical and cell assay data for 18 internal targets used for evaluation only, not new wet-lab work",
    "source_url": "https://arxiv.org/html/2602.07735"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Proprietary evaluation data (internal target assays)",
    "targets": 18,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "18 internal targets with more than 95 IC50 points each; nine proprietary crystals across two targets used only in a fine-tuning experiment",
    "success_rate": "better than the comparator on 15 of 18 targets (developer-reported)",
    "affinity_best": null,
    "version": "arXiv v1",
    "source_type": "self",
    "source_url": "https://arxiv.org/html/2602.07735",
    "date": {
     "date": "2026-02-08",
     "precision": "day",
     "year_hint": null,
     "basis": "arXiv v1",
     "event_kind": "paper-revision",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2602.07735",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "The data belong to the developer; not an independent evaluation and not a new wet-lab campaign",
    "unverified_marker": null,
    "note": "The data come from ongoing projects, are uncurated and span several chemical series; the total compound count is not given"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "evidence_url": "https://arxiv.org/abs/2602.07735",
    "confidence": "high",
    "second_hand": false,
    "note": "Post-hoc evaluation on a released round's affinity set, not participation"
   },
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2602.07735",
    "confidence": "medium",
    "second_hand": false,
    "note": "Only a bar chart merging three structural sets is given; no standalone value for this benchmark appears in the text"
   }
  ],
  "other_evaluations": [
   {
    "label": "Two structural sets (off-site)",
    "category": "public-benchmark-not-on-site",
    "detail": "Merged into the same chart with two other public structural sets; pose accuracy self-reported as comparable to an earlier baseline",
    "source_type": "self"
   },
   {
    "label": "Affinity evaluation on 18 internal targets",
    "category": "own-task-set",
    "detail": "Per-target correlation comparison on developer-owned data that outsiders cannot recompute",
    "source_type": "self"
   },
   {
    "label": "End-to-end speed comparison",
    "category": "comparison-vs-baseline-only",
    "detail": "A speed-up measured at a fixed GPU, token length and sample count per complex; the conclusion depends on that setup",
    "source_type": "self"
   },
   {
    "label": "Role of a high-throughput activity dataset",
    "category": "comparison-vs-baseline-only",
    "detail": "In the paper that dataset is only a source of affinity training data, not a reported evaluation, so it cannot be set against the baseline's same-named result",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a much smaller trunk with diffusion-free optimisation matches an earlier baseline's pose accuracy, runs more than an order of magnitude faster end to end, and correlates better than the comparator on a released affinity set and on internal targets.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "With no code or weights, no third-party evaluation was found; the claimed advantage partly rests on developer-owned target data",
  "background": {
   "org_profile": "A US AI small-molecule drug-discovery company founded in 2018 (headquartered in California; the paper's address differs from the city in earlier coverage), running its own pipeline alongside pharma collaborations; its platform includes a data engine the company says produces a billion-scale set of small-molecule binding measurements with ultra-miniaturised chips.",
   "funding": "A $60M Series A in 2022 led by one venture firm; a $120M Series B announced 2024-10-17 co-led by a new investor and a chipmaker's venture arm with several other participants; the company reports over $200M raised in total.",
   "products_partnerships": "Collaborations include a large pharma company (2023-12, multi-target), a life-science research company, a transcription-factor-focused biotech (2024-10) and another large pharma company (2024-12); public materials mention upfronts, milestones and tiered royalties with no amounts disclosed.",
   "related_models": [
    "model-boltz-2",
    "model-nesso-1",
    "model-aqaffinity"
   ]
  },
  "reading_notes": [
   "A substantial part of the claimed affinity advantage comes from developer-owned targets that outsiders cannot reproduce.",
   "The speed claim depends on a specific hardware and sampling setup and need not hold elsewhere.",
   "PoseBusters appears only in a merged bar chart with no standalone value, so it cannot be compared with same-named figures elsewhere.",
   "Other teams are following the diffusion-free slimming route, suggesting affinity tasks may not need full atom-level diffusion."
  ],
  "caveats": [
   "The latest update is a company press release (promotional), not a model version, and must display as unverified.",
   "No public code or weights, so the results cannot be independently recomputed.",
   "Parameter comparisons hold only at the trunk level.",
   "The CASP entry is retrospective dataset use and does not count toward reported benchmarks."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2602.07735",
    "title": "arXiv 2602.07735 v1",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2602.07735",
    "title": "Paper full text (HTML)",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20260212470275/en",
    "title": "Commercial launch press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://terraytx.com/insights/emmi-predict-terrabind",
    "title": "Company product page",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2026-02-12 recorded as a model update",
    "now": "Same date, but it is a press release and the kind is no later sourced event; must be marked unverified",
    "source_url": "https://www.businesswire.com/news/home/20260212470275/en",
    "date_found": "2026-11-16"
   },
   {
    "field": "other_evaluations",
    "was": "Reports evaluation results on a high-throughput activity dataset",
    "now": "In the paper this dataset is only a training-data source, with no evaluation results",
    "source_url": "https://arxiv.org/html/2602.07735",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "release_milestones.official_announcement",
   "background.funding",
   "background.products_partnerships"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-terrabind/",
   "en": "/bio/benchmark/models/terrabind/"
  },
  "names_full_null_reason": "The paper gives no expanded name; the commercial product name is listed as an alias",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1,
    "retrospective-dataset-use": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-tfold",
  "kind": "model",
  "short_slug": "tfold",
  "names": {
   "display": "tFold",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "tFold",
    "tFold-Ab",
    "tFold-Ag"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "antibody",
  "domains_secondary": [],
  "one_liner": "Antibody structure and design system: MSA-free structure prediction plus epitope-specific de novo design",
  "org": {
   "names": [
    "Tencent AI for Life Sciences Lab"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Antibody structure and design system",
  "open_status": {
   "code": "partial",
   "weights": "on-request",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "GitHub states source and weights are provided but declares no standard open-source licence"
  },
  "first_release": {
   "date": "2024-02-08",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "bioRxiv v1 preprint (repository creation date not used)",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.02.05.578892",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2024-02-08",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "bioRxiv v1",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.02.05.578892",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-12-10",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Journal online publication (design pipeline)",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41467-025-67361-9",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2025-03-14",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "GitHub v1.0 packaging release",
    "timestamp_utc": null,
    "source_url": "https://github.com/TencentAI4S/tfold",
    "verification": "verified",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-12-10",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Journal online publication of the design pipeline",
   "description": "The design-pipeline paper was published; the last GitHub push is the 2025-03-14 packaging release",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1038/s41467-025-67361-9",
   "verification": "verified",
   "note": null
  },
  "status": "active",
  "status_note": "Journal publication within 12 months",
  "what_it_does": "An antibody structure and design system: predicts antibody structure without MSA, predicts antibody-antigen complexes, and combines virtual screening with epitope-specific de novo design; the online service sits on the company's drug-discovery platform.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public antibody structure data (2022-07 time split) and public sequences for the language model",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "not-disclosed",
    "detail": "Own wet-lab training data not disclosed",
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Phage display and SPR validated four targets",
    "source_url": null
   }
  },
  "wetlab_validation": [
   {
    "campaign": "De novo design, four targets",
    "modality": "Fab",
    "targets": 4,
    "designs_tested": 10,
    "hits": null,
    "denominator_basis": "10,000 Fab pairs synthesised per target, with 10 Fc-fusion antibodies finally measured by SPR; success about 1.5%–4.4% with no per-target breakdown",
    "success_rate": "approximate success rate 1.5-4.4% (no per-target breakdown)",
    "affinity_best": "best PD-L1 KD 45 pM; best PD-1 KD 81.6 nM",
    "version": "Nature Communications paper",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12815924/",
    "date": null,
    "verification": "verified",
    "note": "Verified in the published paper, lifting the earlier unverified flag; the site's old \"~82 nM\" should read 81.6 nM. The preprint previously cited as a tFold source actually belongs to IgGM, whose figures must not be attributed here"
   }
  ],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Time-split antibody structure test",
    "category": "own-task-set",
    "detail": "169 time-split antibody structures compared with general folding and antibody-specific models",
    "source_type": "self"
   },
   {
    "label": "Antibody-antigen docking test",
    "category": "own-task-set",
    "detail": "Docking success on 99 complexes; alone it trails a general co-folding model, and improves with ensembling or epitope input",
    "source_type": "self"
   },
   {
    "label": "Virtual screening",
    "category": "own-task-set",
    "detail": "Candidate pools for two targets ranked, with enrichment metrics reported by the developers",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report MSA-free antibody structure prediction better than general folding models on antibody regions, and SPR-confirmed de novo binders on four targets",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent evaluation found",
  "background": {
   "org_profile": "An in-house lab in Shenzhen funded by the company, with no separate funding",
   "funding": null,
   "funding_null_reason": "Not disclosed or not applicable",
   "products_partnerships": "The online service sits on the company's drug-discovery platform; the sister system IgGM shares authors",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-iggm",
    "model-rfantibody",
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "The site once listed the IgGM preprint as a source here; the figures must be kept apart",
   "Only the overall 1.5%–4.4% range exists, with no per-target numbers",
   "No results reported on any of the 44 benchmarks"
  ],
  "caveats": [
   "No standard open-source licence is declared and the weight terms are unclear",
   "Structure and docking evaluations are self-built time-split sets"
  ],
  "sources": [
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12815924/",
    "title": "PubMed/PMC record",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2024.02.05.578892",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://api.crossref.org/works/10.1038/s41467-025-67361-9",
    "title": "Crossref record 10.1038/s41467-025-67361-9",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/TencentAI4S/tfold",
    "title": "GitHub repository TencentAI4S/tfold",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41467-025-67361-9",
    "title": "Journal article DOI 10.1038/s41467-025-67361-9",
    "kind": "journal",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "sources",
    "was": "Listed the IgGM preprint as a tFold source",
    "now": "Changed to this model's journal paper; IgGM figures must not be attributed here",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12815924/",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "Best PD-1 KD about 82 nM",
    "now": "81.6 nM (published paper)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12815924/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-tfold/",
   "en": "/bio/benchmark/models/tfold/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-transcriptformer",
  "kind": "model",
  "short_slug": "transcriptformer",
  "names": {
   "display": "TranscriptFormer",
   "full": null,
   "aliases": [
    "TranscriptFormer",
    "TF-Metazoa",
    "TF-Exemplar",
    "TF-Sapiens"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Variants are treated as versions within one record",
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "CZI/Biohub cross-species generative single-cell model trained on about 110M cells from 12 species",
  "org": {
   "names": [
    "Chan Zuckerberg Initiative",
    "Chan Zuckerberg Biohub"
   ],
   "global_entity_ids": [
    "cz-biohub"
   ]
  },
  "org_type": "nonprofit-institute",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Single-cell FM (generative)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Code and weights released via PyPI and a public data registry under MIT (from the registry page and a second-hand source)"
  },
  "first_release": {
   "date": "2025-04-29",
   "precision": "day",
   "year_hint": null,
   "basis": "bioRxiv preprint posted (registry record)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1101/2025.04.25.650731",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-04-29",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint posted",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2025.04.25.650731",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-04-29",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1101/2025.04.25.650731",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2026-07-02",
    "precision": "day",
    "year_hint": null,
    "basis": "Science publication (registry record)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1126/science.aec8514",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-07-02",
   "precision": "day",
   "year_hint": null,
   "basis": "Science publication",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://api.crossref.org/works/10.1126/science.aec8514",
   "verification": "verified",
   "description": "Journal version published in Science; GitHub v0.6.1 had been released on 2025-11-06",
   "note": null
  },
  "status": "active",
  "status_note": "Journal paper and code release within 12 months",
  "what_it_does": "Models each cell's transcriptome autoregressively as a \"cell sentence\", trained across 12 species; it supports cross-species cell-type transfer, zero-shot disease-state detection and generation of cell expression, with no perturbation evaluation. Unlike C2S-Scale, which turns expression into text and fine-tunes a general LLM, it is a dedicated single-cell autoregressive model.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Mainly CELLxGENE, about 112M cells across 12 species",
    "source_url": "https://doi.org/10.1101/2025.04.25.650731"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1101/2025.04.25.650731"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://doi.org/10.1101/2025.04.25.650731"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Cross-species cell-type classification",
    "category": "own-task-set",
    "detail": "Linear probes on 5 out-of-distribution species against UCE, ESM2-CE, scGPT and Geneformer; author-reported",
    "source_type": "self"
   },
   {
    "label": "Spermatogenesis cross-species mapping",
    "category": "own-task-set",
    "detail": "Label transfer across 9 vertebrate species; author-reported",
    "source_type": "self"
   },
   {
    "label": "Human cell type and disease state",
    "category": "own-task-set",
    "detail": "Tabula Sapiens 2.0 across 26 tissues plus infected vs uninfected lung cells; author-reported",
    "source_type": "self"
   },
   {
    "label": "scIB",
    "category": "public-benchmark-not-on-site",
    "detail": "Not a pairing: scIB-family data are used but no scIB metrics are reported",
    "source_type": "self"
   },
   {
    "label": "Gene-gene and TF analyses",
    "category": "case-study",
    "detail": "Qualitative recovery of known interactions against STRING and cell-type TF prediction",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: a generative model pretrained on about 112M cells from 12 species transfers cell-type labels across roughly 685 million years of divergence and detects disease states zero-shot, beating UCE and other baselines.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No independent third-party evaluation of its cross-species results was found",
  "background": {
   "org_profile": "The Chan Zuckerberg Initiative is a philanthropy that in 2025-11 put Biohub at the centre of its biology and AI work, with CZI as supporting infrastructure; Biohub has sites in San Francisco, New York and Chicago, and the EvolutionaryScale team has joined it.",
   "funding": "Founders' personal philanthropic funds; Biohub says it plans to reach 10,000 GPUs by 2028, with no AI-specific amount published.",
   "products_partnerships": "TranscriptFormer is the first step of CZI's virtual-cell strategy; its data source CELLxGENE is shared with STATE, Stack and Tahoe-x1.",
   "related_models": [
    "model-c2s-scale",
    "model-state",
    "model-stack",
    "model-tahoe-x1",
    "model-scgpt-geneformer"
   ]
  },
  "reading_notes": [
   "With no scIB or perturbation scores, it cannot be compared directly with Geneformer, scGPT or STATE.",
   "Its strength is cross-species transfer, a task without a shared third-party board.",
   "Training is all public data, and its value for drug discovery is untested in the wet lab."
  ],
  "caveats": [
   "All evaluations are the authors' own task sets with no third-party reproduction.",
   "Site figures come from the bioRxiv text; the Science version is paywalled and takes precedence.",
   "Licence information partly comes from a second-hand source."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1126/science.aec8514",
    "title": "Science paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://doi.org/10.1101/2025.04.25.650731",
    "title": "bioRxiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/czi-ai/transcriptformer/releases",
    "title": "Release list",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://pypi.org/project/transcriptformer/",
    "title": "Package registry page",
    "kind": "package-registry",
    "date": null
   },
   {
    "url": "https://registry.opendata.aws/biohub-transcriptomics-mit/",
    "title": "Public data registry page",
    "kind": "official-site",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-transcriptformer/",
   "en": "/bio/benchmark/models/transcriptformer/"
  },
  "names_full_null_reason": "The name is itself the full name",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {}
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developing institute but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-txagent",
  "kind": "model",
  "short_slug": "txagent",
  "names": {
   "display": "TxAgent",
   "full": null,
   "full_null_reason": "No separate expanded name",
   "aliases": [
    "TxAgent"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "Single-model entry, not a family",
  "domain": "text-agents",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Therapeutic-reasoning agent calling 211 drug and clinical tools via ToolUniverse",
  "org": {
   "names": [
    "Harvard Medical School"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Therapeutic-reasoning agent",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Open source; pip packages txagent and tooluniverse"
  },
  "first_release": {
   "date": "2025-03-14",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "arXiv 2503.10970 v1 submitted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2503.10970",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": null,
   "paper_v1": {
    "date": "2025-03-14",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "arXiv v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2503.10970",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-04-21",
   "precision": "day",
   "event_kind": "none-since-release",
   "basis": "Last TxAgent repository commit (v0.1.3)",
   "description": "No later sourced event found (as of 2026-11-16)",
   "timestamp_utc": null,
   "source_url": "https://github.com/mims-harvard/TxAgent",
   "verification": "verified",
   "note": "The separate ToolUniverse package is still releasing (v1.5.10 on 2026-11-14); not a TxAgent update"
  },
  "status": "maintained",
  "status_note": "No TxAgent commits after 2025-04",
  "what_it_does": "A therapeutics reasoning agent that picks among 211 drug and clinical tools via ToolUniverse to answer questions on drug interactions, pharmacokinetics and regulatory status through multi-step reasoning.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Public sources behind the 211 tools, e.g. FDA drug information and Open Targets",
    "source_url": null
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [],
  "other_evaluations": [
   {
    "label": "Five self-built benchmarks",
    "category": "own-task-set",
    "detail": "DrugPC, BrandPC, GenericPC, TreatmentPC, DescriptionPC: 3,168 drug tasks plus 456 treatment scenarios",
    "source_type": "self"
   },
   {
    "label": "Tool-use ablation",
    "category": "comparison-vs-baseline-only",
    "detail": "ToolRAG ablation over 211 tools, compared with general LLM baselines",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report higher open-ended drug-reasoning accuracy than GPT-4o and DeepSeek-R1 and generalization across drug-name variants",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Only CURE-Bench, built by the same lab",
  "background": {
   "org_profile": "Academic lab at Harvard Medical School (also tied to the Kempner Institute and Broad Institute), with MIT Lincoln Laboratory collaborators",
   "funding": "Acknowledged funding from the Milken Institute, HMS Dean's Innovation Fund and the Kempner Institute",
   "funding_null_reason": null,
   "products_partnerships": "Companion tool ecosystem ToolUniverse; the same lab built CURE-Bench (NeurIPS 2025)",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-biomni",
    "model-txgemma"
   ]
  },
  "reading_notes": [
   "All evaluations are developer-built task sets outside the 44 benchmarks",
   "Still on the zero-pairing list under this round's relation rules"
  ],
  "caveats": [
   "Accuracy is developer-reported only; per-task breakdown only in the full text",
   "No wet-lab validation"
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2503.10970",
    "title": "arXiv preprint 2503.10970",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/mims-harvard/TxAgent",
    "title": "GitHub repository mims-harvard/TxAgent",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/mims-harvard/ToolUniverse",
    "title": "GitHub repository mims-harvard/ToolUniverse",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://zitniklab.hms.harvard.edu/TxAgent",
    "title": "Official page zitniklab.hms.harvard.edu",
    "kind": "company-blog",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-txagent/",
   "en": "/bio/benchmark/models/txagent/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-txgemma",
  "kind": "model",
  "short_slug": "txgemma",
  "names": {
   "display": "TxGemma",
   "full": null,
   "aliases": [
    "TxGemma",
    "txgemma-9b-predict"
   ],
   "ambiguous_aliases": [
    "Gemma"
   ]
  },
  "family": "Gemma",
  "family_null_reason": null,
  "domain": "small-molecule",
  "domains_secondary": [
   "text-agents"
  ],
  "one_liner": "A family of drug-discovery LLMs (three sizes, predict and chat variants) instruction-tuned from an open general base on a public therapeutics dataset library",
  "org": {
   "names": [
    "Google DeepMind",
    "Google Research"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "GB",
   "US"
  ],
  "model_class": "Drug-discovery LLM (instruction-tuned)",
  "open_status": {
   "code": "partial",
   "weights": "research-only",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Weights are open subject to the developer's health-AI terms; the released version was trained only on commercially licensed data, unlike the paper version"
  },
  "first_release": {
   "date": "2025-03-25",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer blog announcing the open weights",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://developers.googleblog.com/introducing-txgemma-open-models-improving-therapeutics-development/",
   "verification": "verified",
   "note": "The model hub repository creation date of 2025-03-21 is not used as first release under this round's rule; the paper v1 is 2025-04-08"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-03-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://developers.googleblog.com/introducing-txgemma-open-models-improving-therapeutics-development/",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2025-04-08",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint v1",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2504.06196",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": {
    "date": "2025-03-25",
    "precision": "day",
    "year_hint": null,
    "basis": "All three sizes released with the announcement, subject to the developer's terms",
    "event_kind": "weights-release",
    "timestamp_utc": null,
    "source_url": "https://developers.google.com/health-ai-developer-foundations/txgemma/model-card",
    "verification": "verified",
    "note": null
   },
   "challenge_open": null,
   "official_announcement": {
    "date": "2025-03-25",
    "precision": "day",
    "year_hint": null,
    "basis": "Official blog announcing open weights",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://developers.googleblog.com/introducing-txgemma-open-models-improving-therapeutics-development/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2025-04-10",
   "precision": "day",
   "year_hint": null,
   "basis": "Last modification date of the three model hub repositories",
   "event_kind": "code-maintenance-only",
   "timestamp_utc": null,
   "source_url": "https://huggingface.co/google/txgemma-9b-predict",
   "verification": "verified",
   "description": "That date is a repository modification, not a new model version; no later TxGemma version was found as of 2026-11-16",
   "note": null
  },
  "status": "maintained",
  "status_note": "No new version within 12 months, only repository modifications",
  "what_it_does": "Rewrites the tasks of a public therapeutics dataset library as instruction data and fine-tunes an open general base so that one model covers classification, regression and generation across small-molecule and protein therapeutic entities; the predict variant gives structured output, the chat variant natural-language explanation, and the paper adds an agent form. It is the open successor to the same organisation's earlier closed drug LLM and is often used as a comparator in papers such as Enchant and ether0.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "66 tasks and over 15 million data points from a public therapeutics dataset library",
    "source_url": "https://arxiv.org/abs/2504.06196"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The public weights use only commercially licensed datasets; the paper version also used non-commercial ones",
    "source_url": "https://developers.google.com/health-ai-developer-foundations/txgemma/model-card"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2504.06196"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2504.06196"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2504.06196",
    "confidence": "high",
    "second_hand": false,
    "note": "The paper reports results on 66 tasks of that library under its recommended splits, counted separately against specialist and against general-purpose comparators"
   },
   {
    "benchmark_id": "bench-04-tdc",
    "relation_type": "third-party-evaluation",
    "source_type": "third",
    "evidence_url": "https://iambic.ai/post/enchantv2",
    "confidence": "medium",
    "second_hand": false,
    "note": "A company ran the mid-size variant on its kinase task set as a comparator for its own model; the evaluation set is not that library's official split"
   }
  ],
  "other_evaluations": [
   {
    "label": "Agent-form task demonstrations",
    "category": "own-task-set",
    "detail": "The paper wires the model to tools for multi-step task demonstrations, author-defined and outside the benchmarks on this site",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that a single open-weights model covers most tasks of that dataset library, matching or beating specialist models on most of them.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A company ran it as a comparator on its own kinase task set: a third-party evaluation, though not on an official split.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "The developer is a large technology company's AI research unit, founded in London in 2010, acquired in 2014 and merged with the company's other research team in 2023-04; this model was released through its health-AI developer programme as internal research.",
   "funding": null,
   "products_partnerships": "Its predecessor was the same organisation's closed drug LLM on a different base; the base here is the company's open general model; related structural-biology work is AlphaFold 3.",
   "related_models": [
    "model-alphafold3",
    "model-enchant-v1-v2",
    "model-ether0"
   ]
  },
  "reading_notes": [
   "Its selling point is covering many tasks in one model, not topping any single task.",
   "The paper keeps separate counts against specialist and against general-purpose comparators; the two must not be merged when quoted.",
   "The paper version and the public weights were trained on different data, so reproducing the paper with the open weights may fall short.",
   "Training and evaluation come from the same dataset library, so transfer to real projects is unverified."
  ],
  "caveats": [
   "First release follows the official blog; the model hub's repository creation date is not used as evidence.",
   "2025-04-10 is a repository modification date and must not be presented as a new version.",
   "The third-party kinase task set is not that library's official split, so its figures cannot be listed beside official-protocol ones."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2504.06196",
    "title": "Paper preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://developers.google.com/health-ai-developer-foundations/txgemma/model-card",
    "title": "Model card",
    "kind": "official-page",
    "date": null
   },
   {
    "url": "https://developers.googleblog.com/introducing-txgemma-open-models-improving-therapeutics-development/",
    "title": "Launch blog",
    "kind": "blog",
    "date": "2025-03-25"
   },
   {
    "url": "https://huggingface.co/google/txgemma-9b-predict",
    "title": "Model hub repository",
    "kind": "model-hub",
    "date": null
   },
   {
    "url": "https://iambic.ai/post/enchantv2",
    "title": "Third-party evaluation post",
    "kind": "blog",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "No later update (N/A)",
    "now": "Repository change on 2025-04-10, kind recorded as maintenance only, not a new version",
    "source_url": "https://huggingface.co/google/txgemma-9b-predict",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2025-03 (from memory, not sourced)",
    "now": "2025-03-25, based on the official release blog",
    "source_url": "https://developers.googleblog.com/introducing-txgemma-open-models-improving-therapeutics-development/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-txgemma/",
   "en": "/bio/benchmark/models/txgemma/"
  },
  "names_full_null_reason": "The name is a product name; the model card gives no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1,
    "third-party-evaluation": 1
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-uni-mol2",
  "kind": "model",
  "short_slug": "uni-mol2",
  "names": {
   "display": "Uni-Mol2",
   "full": null,
   "aliases": [
    "Uni-Mol2"
   ],
   "ambiguous_aliases": [
    "Uni-Mol"
   ]
  },
  "family": null,
  "family_null_reason": "Each generation has its own record; the bare name without a digit is ambiguous and is never auto-linked",
  "domain": "small-molecule",
  "domains_secondary": [],
  "one_liner": "3D-conformation molecular pretraining model scaled to the billion-parameter range to study scaling",
  "org": {
   "names": [
    "DP Technology",
    "Peking University",
    "AI for Science Institute, Beijing"
   ],
   "global_entity_ids": [
    "dp-technology"
   ]
  },
  "org_type": "platform-company",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Property prediction (3D-conformation molecular pretraining)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": "MIT",
   "license_changed": null,
   "note": "The repository is MIT overall and the weight sizes are public on the host; no separate weights licence is listed"
  },
  "first_release": {
   "date": "2024-06-21",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2406.14969 v1 submitted (v2 on 2024-07-01)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2406.14969",
   "verification": "verified",
   "note": "The site previously gave only 2024-06"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-06-21",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2406.14969",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-06-21",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2406.14969",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2024-10-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Pull request adding the model codebase merged into the main repository",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/deepmodeling/Uni-Mol",
    "verification": "partial",
    "note": "The site said a 2024-10-10 release tag; no commit exists that day and the releases list ends at v0.2.1 (2024-07-06)"
   },
   "weights_release": {
    "date": "2024-10-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Weights repository created with several model sizes uploaded (84M up to the billion range)",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://huggingface.co/dptech/Uni-Mol2",
    "verification": "partial",
    "note": "Relies on the hosting site's repository creation date; marked unverified"
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2024-10-18",
   "precision": "day",
   "year_hint": null,
   "basis": "Pull request adding the model code to the main repository",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/deepmodeling/Uni-Mol",
   "verification": "partial",
   "description": "Code and several pretrained weight sizes landed in the main repository and the weights host (weights repo created 2024-10-17); the paper was also accepted at a 2024 machine-learning conference; commits through 2025-05 are maintenance. This is the latest sourced substantive event",
   "note": "The earlier 2024-10-10 release-tag framing is corrected"
  },
  "status": "maintained",
  "status_note": "Latest substantive event in 2024-10, with maintenance commits only afterwards",
  "what_it_does": "Pretrains on molecules represented as 3D conformations and pushes both parameter and conformation scale, reporting how size and data relate to downstream performance; the paper positions itself as a scaling study of molecular pretraining. Compared with MiniMol and MolE it needs pregenerated conformations, is two orders of magnitude larger, and reports quantum-chemistry property tasks rather than drug ADMET; the sibling successor Uni-Mol3 targets reactions.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 19 million molecules from the predecessor plus a subset of a public compound library (about 884 million compounds), about 800 million conformations in total",
    "source_url": "https://arxiv.org/abs/2406.14969"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The paper reports no wet-lab validation",
    "source_url": "https://arxiv.org/abs/2406.14969"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-03-moleculenet",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2406.14969v2",
    "confidence": "medium",
    "second_hand": false,
    "note": "Only one quantum-chemistry property set within the benchmark is reported (scaffold split); the other datasets are not"
   }
  ],
  "other_evaluations": [
   {
    "label": "Off-site quantum-chemistry property dataset",
    "category": "public-benchmark-not-on-site",
    "detail": "A property dataset of several thousand molecules compared with the predecessor, in settings with and without atom and bond features; developer-reported",
    "source_type": "self"
   },
   {
    "label": "Own scaling study",
    "category": "own-task-set",
    "detail": "A self-built comparison across model sizes and data volumes supporting the scaling claim, with tasks chosen by the authors",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: scaling parameters and conformations together improves quantum-chemistry property tasks over the predecessor and yields a scaling relationship for molecular pretraining.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation of this model on a site benchmark was found",
  "background": {
   "org_profile": "The developer is an AI-for-Science tooling and platform company founded in Beijing in 2018, with a research cloud platform and laboratory and drug-design platforms; collaborators are a Chinese university and a Beijing AI-for-science institute.",
   "funding": "A Series C of over RMB 800 million (about US$114 million) was announced 2025-12-24, led mainly by industrial and municipal funds; earlier rounds appear in the same company's other records.",
   "products_partnerships": "The model series is open-sourced under the same organisation's public repositories alongside platform and tooling products; this model's goal is to test the scaling relationship of molecular pretraining.",
   "related_models": [
    "model-uni-mol3",
    "model-minimol",
    "model-mole",
    "model-kermt"
   ]
  },
  "reading_notes": [
   "The reported tasks are quantum-chemistry properties, a different track from the ADMET competitors on TDC, so the two should not be compared directly.",
   "The scaling conclusion rests on author-chosen tasks; the benefit of scale on small ADMET datasets has no independent verification.",
   "Conformations must be generated in advance, making inference costlier than graph or string models.",
   "The weights are public, making it one of the few large molecular pretraining models outsiders can evaluate themselves; only a single subset of MoleculeNet is reported."
  ],
  "caveats": [
   "Only one subset of the benchmark is reported; the other datasets are not, so coverage is limited.",
   "The latest substantive event is 2024-10 and the weights date relies on a host-side creation date.",
   "No separate weights licence is listed in the repository.",
   "No wet-lab validation and no third-party evaluation."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2406.14969",
    "title": "Model paper, arXiv 2406.14969",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2406.14969v2",
    "title": "Paper v2 full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/deepmodeling/Uni-Mol",
    "title": "Series code repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://huggingface.co/dptech/Uni-Mol2",
    "title": "Weights hosting repository",
    "kind": "package-registry",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "latest_update",
    "was": "2024-10-10 (code and weights release tag)",
    "now": "2024-10-18 (PR merging the model code into the main repository); weights repository created 2024-10-17; the releases list ends at v0.2.1 (2024-07-06)",
    "source_url": "https://github.com/deepmodeling/Uni-Mol",
    "date_found": "2026-11-16"
   },
   {
    "field": "release_milestones.journal_online",
    "was": "No conference or journal version seen",
    "now": "The paper was accepted at a 2024 machine-learning conference (no journal version)",
    "source_url": "https://arxiv.org/abs/2406.14969",
    "date_found": "2026-11-16"
   },
   {
    "field": "first_release",
    "was": "2024-06",
    "now": "2024-06-21（arXiv v1）",
    "source_url": "https://arxiv.org/abs/2406.14969",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "release_milestones.code_release",
   "release_milestones.weights_release",
   "open_status.weights_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-uni-mol2/",
   "en": "/bio/benchmark/models/uni-mol2/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developer company but none for this model",
  "status_head": null
 },
 {
  "id": "model-uni-mol3",
  "kind": "model",
  "short_slug": "uni-mol3",
  "names": {
   "display": "Uni-Mol3",
   "full": null,
   "aliases": [
    "Uni-Mol3"
   ],
   "ambiguous_aliases": [
    "Uni-Mol"
   ]
  },
  "family": null,
  "family_null_reason": "Each generation has its own record; the bare name without a digit is ambiguous and is never auto-linked",
  "domain": "small-molecule",
  "domains_secondary": [
   "generation"
  ],
  "one_liner": "Multi-molecular reaction foundation model covering products, retrosynthesis, conditions and yield",
  "org": {
   "names": [
    "DP Technology",
    "Peking University"
   ],
   "global_entity_ids": [
    "dp-technology"
   ]
  },
  "org_type": "platform-company",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Reaction model / retrosynthesis (multi-molecular foundation model)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": "MIT (per repository)",
   "code_license": "MIT",
   "license_changed": null,
   "note": "Code and weights are MIT per the repository; the licensed patent-reaction datasets cannot be redistributed and only preprocessing scripts are provided"
  },
  "first_release": {
   "date": "2025-07-30",
   "precision": "day",
   "year_hint": null,
   "basis": "arXiv 2508.00920 v1 submitted (v2 on 2025-08-12)",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2508.00920",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2025-07-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Initial code commit (v1.0), earlier than the paper's v1",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/LirongWu/Uni-Mol3",
    "verification": "partial",
    "note": "Date taken from commit history and marked unverified; first_release stays on the paper v1 because the commit precedes any public announcement and has no independent source"
   },
   "paper_v1": {
    "date": "2025-07-30",
    "precision": "day",
    "year_hint": null,
    "basis": "arXiv v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2508.00920",
    "verification": "verified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2025-07-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Initial repository code v1.0",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/LirongWu/Uni-Mol3",
    "verification": "partial",
    "note": null
   },
   "weights_release": {
    "date": "2026-03-24",
    "precision": "day",
    "year_hint": null,
    "basis": "Repository commit adding weight download links",
    "event_kind": "new-weights",
    "timestamp_utc": null,
    "source_url": "https://github.com/LirongWu/Uni-Mol3",
    "verification": "partial",
    "note": "The weights event is inferred from the commit message and unverified"
   },
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-03-24",
   "precision": "day",
   "year_hint": null,
   "basis": "Repository commit adding weight download links",
   "event_kind": "code-release",
   "timestamp_utc": null,
   "source_url": "https://github.com/LirongWu/Uni-Mol3",
   "verification": "partial",
   "description": "A repository commit added download links for pretrained and fine-tuned weights; initial code dates to 2025-07-18 (v1.0). No journal version was found. This is the latest sourced event",
   "note": "The nature of the event is inferred from the commit message"
  },
  "status": "active",
  "status_note": "Substantive repository commit within 12 months",
  "what_it_does": "Treats molecules and reactions as multi-molecular systems, pretraining on molecules and then on reactions at scale, with one model covering forward product prediction, retrosynthesis, condition recommendation and yield prediction. Unlike the reasoning-style retrosynthesis of RetroDFM-R it uses representation pretraining plus reaction fine-tuning; sequence models such as Chemformer are its main retrosynthesis comparison, and Uni-Mol2 is the sibling predecessor.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 19 million molecules for pretraining from public compound and bioactivity libraries; the paper's description of the sources is internally inconsistent",
    "source_url": "https://arxiv.org/html/2508.00920v2"
   },
   "b_private_licensed": {
    "value": "yes",
    "detail": "A commercially licensed patent-reaction database, about 16.67 million records reduced to 11,973,789 after cleaning, not redistributable under the licence",
    "source_url": "https://arxiv.org/html/2508.00920v2"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": "The paper reports no wet-lab validation",
    "source_url": "https://arxiv.org/html/2508.00920v2"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-07-uspto",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/html/2508.00920v2",
    "confidence": "high",
    "second_hand": false,
    "note": "Three official subsets - product prediction, retrosynthesis and reaction conditions - all developer-reported; comparator figures in the same table are the authors' own runs"
   }
  ],
  "other_evaluations": [
   {
    "label": "Retrosynthesis test set inside the licensed database",
    "category": "own-task-set",
    "detail": "Developer-reported retrosynthesis on a test split of the licensed patent-reaction database; outsiders cannot obtain the data, so it is not reproducible",
    "source_type": "self"
   },
   {
    "label": "Reaction condition and yield tasks",
    "category": "own-task-set",
    "detail": "Self-built multi-task evaluation covering condition recommendation and yield prediction",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one model unifies product prediction, retrosynthesis, conditions and yield, with molecule pretraining plus large-scale reaction pretraining giving broadly usable performance.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation of this model on a site benchmark was found",
  "background": {
   "org_profile": "The developer is an AI-for-Science platform company founded in Beijing in 2018, working across drug design, materials and battery research; the collaborator is a mathematics and machine-learning unit at a Chinese university. No citable headcount exists in public material, so none is given.",
   "funding": "Series A of tens of millions of dollars in 2021-08; Series B of tens of millions in 2022-01; over RMB 700 million in 2023-08; a Series C of over RMB 800 million (about US$110 million) on 2025-12-24 with industrial and municipal funds. Outlets name the middle rounds inconsistently, so only years and amounts are recorded.",
   "products_partnerships": "Products include a research cloud platform, a drug-design platform and docking and free-energy tooling; the model series is open-sourced in public repositories. This model's reaction pretraining depends on commercially licensed data that outsiders cannot reproduce.",
   "related_models": [
    "model-uni-mol2",
    "model-chemformer",
    "model-retrodfm-r"
   ]
  },
  "reading_notes": [
   "Its self-reported retrosynthesis result sits below Chemformer in the same table, so the selling point is multi-task unification rather than a single best score (no figures are transcribed here).",
   "Reaction pretraining uses a commercially licensed database that outsiders cannot reproduce; only code and weights are open.",
   "The USPTO test sets suffer template leakage, which limits what high scores mean.",
   "The nature of the latest update is inferred from a commit message; the weights and code axes should be read separately."
  ],
  "caveats": [
   "All benchmark results are developer-reported; there is no board entry and no third-party evaluation.",
   "The paper's description of the public portion of the training data is internally inconsistent.",
   "The licensed reaction database cannot be redistributed, so pretraining is not reproducible.",
   "Both the weights-link event and the journal status are unverified."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2508.00920",
    "title": "Model paper, arXiv 2508.00920",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2508.00920v2",
    "title": "Paper v2 full text",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/LirongWu/Uni-Mol3",
    "title": "Code and weights repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "latest_update",
   "release_milestones.code_release",
   "release_milestones.weights_release",
   "release_milestones.first_public_evidence",
   "training_data.a_public"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-uni-mol3/",
   "en": "/bio/benchmark/models/uni-mol3/"
  },
  "names_full_null_reason": "The paper uses the model name only; no expanded name",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-official-protocol": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developer company but none for this model",
  "status_head": null
 },
 {
  "id": "model-venusrem",
  "kind": "model",
  "short_slug": "venusrem",
  "names": {
   "display": "VenusREM",
   "full": null,
   "full_null_reason": "The paper uses the short name only",
   "aliases": [
    "VenusREM"
   ],
   "ambiguous_aliases": [
    "Venus",
    "VenusREM2"
   ]
  },
  "family": null,
  "family_null_reason": "The successor VenusREM2 and its variants appear only in repository news and are not merged in",
  "domain": "protein",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "Retrieval-augmented zero-shot mutation-effect model fusing sequence, structure and homologs",
  "org": {
   "names": [
    "Shanghai Jiao Tong University",
    "East China University of Science and Technology",
    "Shanghai AI Laboratory"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Protein language model (retrieval-augmented zero-shot)",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Code in the official repository and weights in a public model hub; licence text not retrieved this round, so terms are left empty"
  },
  "first_release": {
   "date": "2024-10-28",
   "precision": "day",
   "event_kind": "preprint-posted",
   "basis": "Preprint v1 posted",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2410.21127",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2024-10-28",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2410.21127",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2024-10-28",
    "precision": "day",
    "event_kind": "preprint-posted",
    "basis": "Preprint v1",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2410.21127",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-07-15",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Journal version online",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1093/bioinformatics/btaf189",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2026-04-25",
    "precision": "day",
    "event_kind": "code-release",
    "basis": "Official repository release v1.0.0",
    "timestamp_utc": null,
    "source_url": "https://github.com/ai4protein/VenusREM",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2026-09-29",
   "precision": "day",
   "event_kind": "successor-model",
   "basis": "The final repository commit announces results for the successor VenusREM2",
   "description": "Repository news announces the successor VenusREM2's board result (README only, unverified); earlier v1.0.0 on 2026-04-25 and journal online 2025-07-15",
   "timestamp_utc": null,
   "source_url": "https://github.com/ai4protein/VenusREM",
   "verification": "partial",
   "note": "The successor's preprint identifier and score appear only in the README and are not independently verified (unverified)"
  },
  "status": "active",
  "status_note": "Repository release and successor news within 12 months",
  "what_it_does": "Builds on the quantized-structure base of ProSST and adds retrieved homologous sequences plus structure information to score point mutations zero-shot, with no DMS labels in training. It is a multimodal sequence-plus-structure-plus-retrieval scorer, a peer of SaProt with an added retrieval path.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "Unsupervised with no DMS training; base data as in ProSST",
    "source_url": "https://github.com/ai4protein/VenusREM"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "No private or licensed data reported",
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "Ten new DNA polymerase mutants assayed for high-temperature activity; a separate batch of antibody mutants is a retrospective analysis",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12261477/"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "DNA polymerase mutants, high-temperature activity",
    "modality": "enzyme",
    "targets": 1,
    "designs_tested": 10,
    "hits": 8,
    "denominator_basis": "The top ten single-point mutants from a saturation scan were assayed (small n, a single enzyme target)",
    "success_rate": "8/10",
    "affinity_best": "best mutant about 8-fold",
    "version": "journal full text",
    "source_type": "self",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12261477/",
    "date": {
     "date": "2025-07-15",
     "precision": "day",
     "event_kind": "journal-publication",
     "basis": "Journal full text online",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1093/bioinformatics/btaf189",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "note": "Developer-reported, not an independent replication: eight mutants significantly improved activity and three improved thermostability and activity together. A separate batch of about 31 antibody mutants is a retrospective analysis of already published data, not a new experiment."
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "counted": true,
    "eval_set": "zero-shot substitution official summary table (Structure & MSA category)",
    "source_url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "confidence": "high",
    "verification": "verified",
    "note": "Listed directly in the official summary table, in the same top tier as another retrieval-augmented model; the paper's state-of-the-art wording is a developer claim"
   }
  ],
  "other_evaluations": [
   {
    "label": "Specialist-predictor control in an external meta-analysis",
    "category": "public-benchmark-not-on-site",
    "detail": "A third-party meta-analysis uses it as a leading specialist predictor against general frontier models, abstract-level material only",
    "source_type": "third"
   },
   {
    "label": "Maintainer blog grouping of multimodal models",
    "category": "comparison-vs-baseline-only",
    "detail": "Board maintainers group it with another multimodal model at the top, a directional observation",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report that retrieval augmentation fused with structure and homolog information gives then-best zero-shot mutation-effect prediction, backed by enzyme-mutant wet-lab work.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "A third-party meta-analysis treats it as a leading specialist predictor against general models (abstract-level material); no independent replication of its wet lab.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "An academic team with no separate company: released in 2025-07 by Shanghai Jiao Tong University with a chemical-technology university and the Shanghai AI Laboratory, largely the same author group as ProSST. The public model-hub entry sits under a research organization account.",
   "funding": null,
   "funding_null_reason": "Paper acknowledgements not retrieved; funding left blank",
   "products_partnerships": "Built on ProSST with retrieval augmentation; the same Venus line includes a protein-engineering system and a mining tool, and the successor VenusREM2 appears only in repository news (unverified).",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-prosst",
    "model-saprot",
    "model-poet-2"
   ]
  },
  "reading_notes": [
   "It shares the official board's top tier with another retrieval-augmented model; differences inside that tier sit at noise level, so do not read rank as a gap.",
   "Retrieval augmentation and structure inputs raise scores markedly, so it is not directly comparable with sequence-only models.",
   "The only new experiment is ten enzyme mutants (8/10 improved activity): a single target with small n, which says nothing about design hit rates.",
   "The roughly 31 antibody mutants are a retrospective analysis and must never be added to the new experiment."
  ],
  "caveats": [
   "The successor VenusREM2's result and preprint identifier appear only in repository news (unverified).",
   "The wet-lab work is developer-reported, not an independent replication.",
   "The paper's state-of-the-art claim and the board's tie at the top are two different statements."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1093/bioinformatics/btaf189",
    "title": "VenusREM journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12261477/",
    "title": "Journal full text, source of the wet-lab figures",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2410.21127",
    "title": "VenusREM preprint v1",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/ai4protein/VenusREM",
    "title": "Official repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "ProteinGym official summary table",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://huggingface.co/OneScience-Group/VenusREM",
    "title": "Public model-hub entry",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "The site said per-mutant data were absent and no hit rate could be assessed",
    "now": "The journal full text reports 8 of 10 mutants significantly improved, the best about 8-fold, and 3 improving thermostability and activity together",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12261477/",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "About 31 antibody mutants were treated as another experiment",
    "now": "Recorded as a retrospective analysis of published data, not a new experiment, and not added to wet-lab scale",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12261477/",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.date",
    "was": "2026-09-30",
    "now": "2026-09-29, the date of the final repository commit",
    "source_url": "https://github.com/ai4protein/VenusREM",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "release_milestones.code_release",
   "background.products_partnerships",
   "open_status.weights_license",
   "open_status.code_license"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-venusrem/",
   "en": "/bio/benchmark/models/venusrem/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-verid",
  "kind": "model",
  "short_slug": "verid",
  "names": {
   "display": "VERID",
   "full": null,
   "aliases": [
    "VERID"
   ],
   "ambiguous_aliases": []
  },
  "family": null,
  "family_null_reason": "A single infrastructure whose two companion applications are described in one record",
  "domain": "text-agents",
  "domains_secondary": [
   "protein"
  ],
  "one_liner": "Semantic digital twin of a lab workstation that loops agents between virtual dry runs, safety checks and real wet-lab experiments",
  "org": {
   "names": [
    "Google Research"
   ],
   "global_entity_ids": []
  },
  "org_type": "big-tech-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Agent infrastructure / lab digital twin",
  "open_status": {
   "code": "unclear",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Built on an open-source lab robotics library; neither the blog nor the paper abstracts state whether its own code or weights are released"
  },
  "first_release": {
   "date": "2026-10-16",
   "precision": "day",
   "year_hint": null,
   "basis": "Official research blog post",
   "event_kind": "official-announcement",
   "timestamp_utc": null,
   "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-10-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Official research blog",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2026-10-23",
    "precision": "day",
    "year_hint": null,
    "basis": "Main preprint; the two companion papers are dated 2026-10-17 and 2026-10-23",
    "event_kind": "preprint-posted",
    "timestamp_utc": null,
    "source_url": "https://arxiv.org/abs/2610.32755",
    "verification": "partial",
    "note": null
   },
   "journal_online": null,
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-10-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Official research blog",
    "event_kind": "official-announcement",
    "timestamp_utc": null,
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "verification": "verified",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-10-23",
   "precision": "day",
   "year_hint": null,
   "basis": "Companion papers posted; no sourced substantive event since",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://arxiv.org/abs/2610.32755",
   "verification": "partial",
   "description": "The main paper and one companion paper appeared the same day as part of the same release rather than a new version; nothing states whether code or weights are released",
   "note": null
  },
  "status": "active",
  "status_note": "First published within 12 months, with companion papers",
  "what_it_does": "Maintains a persistent semantic digital twin of a lab workstation, tracking robot geometry, well volumes, reagent identity and consumable stock, and offering introspection, time-compressed simulation, experiment provenance and a safe execution framework; two companion applications connect agents on top of it to real wet-lab work (three rounds of aptamer design and test, and binder design that trims the target to an interface window to speed design up). It divides work with Co-Scientist, which produces hypotheses while this addresses reliability once a hypothesis meets physical experiments; the companion work's structure and diffusion steps use tools such as AlphaFold 3.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "Training data of the underlying language model and design tools is undisclosed; public structure-database entries serve as design targets",
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/"
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "Experimental feedback of about 1,000 candidates per round and some 4,400 constructs in total tunes design rules online in the aptamer loop; this is not pretraining",
    "source_url": "https://arxiv.org/abs/2610.24806"
   },
   "d_wetlab_validation_only": {
    "value": "yes",
    "detail": "The binder side's yeast display and sorting is validation",
    "source_url": "https://arxiv.org/html/2610.33195v1"
   }
  },
  "wetlab_validation": [
   {
    "campaign": "Aptamer design loop (first target)",
    "targets": 1,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "Bead-based selection with bead-corrected hit rates reported per round; the number tested per round is not given",
    "success_rate": "round 1 0%, round 2 26.0%, round 3 42.1% at 98.5% specificity",
    "affinity_best": null,
    "version": "blog / aptamer paper",
    "source_type": "self",
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "date": {
     "date": "2026-10-16",
     "precision": "day",
     "year_hint": null,
     "basis": "Official blog",
     "event_kind": "official-announcement",
     "timestamp_utc": null,
     "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Designed and read out by the developer; no blind third party",
    "unverified_marker": null,
    "note": "The zero first round is attributed to fixed primer overhangs disrupting folding in nearly half the sequences; the intermediate round was previously missing on the site and all three rounds must be shown"
   },
   {
    "campaign": "Aptamer design loop (second target)",
    "targets": 1,
    "designs_tested": null,
    "hits": null,
    "denominator_basis": "The same loop's second target, reported as specificity paired with hit rate rather than three plain hit rates",
    "success_rate": "round 2 specificity 75.0% to 19.1% with bead-corrected hit rate 10.2% to 3.7%; round 3 specificity 72.4% with on-target hit rate 33.5%",
    "affinity_best": null,
    "version": "blog / aptamer paper",
    "source_type": "self",
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "date": {
     "date": "2026-10-16",
     "precision": "day",
     "year_hint": null,
     "basis": "Official blog",
     "event_kind": "official-announcement",
     "timestamp_utc": null,
     "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
     "verification": "verified",
     "note": null
    },
    "verification": "verified",
    "executor": "Designed and read out by the developer",
    "unverified_marker": "The first-round figure for this target appears only in a chart and is not citable",
    "note": "The second round's drop in specificity came from one structural motif dominating the designs, and the third round recovered through experimental feedback"
   },
   {
    "campaign": "New binder experiment (single target)",
    "targets": 1,
    "designs_tested": 1250,
    "hits": null,
    "denominator_basis": "New experiment: a 5,000-member library with 1,250 designs per arm, read out by yeast display and sorting",
    "success_rate": null,
    "affinity_best": "single-digit nanomolar (top designs)",
    "version": "companion paper",
    "source_type": "self",
    "source_url": "https://arxiv.org/html/2610.33195v1",
    "date": {
     "date": "2026-10-23",
     "precision": "day",
     "year_hint": null,
     "basis": "Companion paper",
     "event_kind": "preprint-posted",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2610.33195",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "Run by the developer",
    "unverified_marker": null,
    "note": "This is the newly run experiment and must not be merged with the computational analysis on existing data below"
   },
   {
    "campaign": "Computational verification on existing data (not wet lab)",
    "targets": 15,
    "designs_tested": 3666,
    "hits": null,
    "denominator_basis": "The 15 targets and 3,666 designs are a trimmed analysis subset of an existing benchmark, not new experimental material",
    "success_rate": "computational hit rate 5.06% to 9.33% at equal compute",
    "affinity_best": null,
    "version": "companion paper",
    "source_type": "self",
    "source_url": "https://arxiv.org/html/2610.33195v1",
    "date": {
     "date": "2026-10-23",
     "precision": "day",
     "year_hint": null,
     "basis": "Companion paper",
     "event_kind": "preprint-posted",
     "timestamp_utc": null,
     "source_url": "https://arxiv.org/abs/2610.33195",
     "verification": "partial",
     "note": null
    },
    "verification": "verified",
    "executor": "A computational analysis with no experimental executor",
    "unverified_marker": null,
    "note": "It must be presented as computational verification on existing data, and these figures must never be read as new wet-lab scale"
   }
  ],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-14-pubmedqa",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "confidence": "high",
    "second_hand": false,
    "note": "No result on that evaluation was found in the official material"
   },
   {
    "benchmark_id": "bench-18-labbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "confidence": "high",
    "second_hand": false,
    "note": "No result on that evaluation was found in the official material"
   },
   {
    "benchmark_id": "bench-23-bioasq",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "confidence": "high",
    "second_hand": false,
    "note": "No result on that evaluation was found in the official material"
   },
   {
    "benchmark_id": "bench-29-bixbench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "confidence": "high",
    "second_hand": false,
    "note": "No result on that evaluation was found in the official material"
   },
   {
    "benchmark_id": "bench-41-bioml-bench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "confidence": "high",
    "second_hand": false,
    "note": "No result on that evaluation was found in the official material"
   },
   {
    "benchmark_id": "bench-42-biomysterybench",
    "relation_type": "disputed-or-unconfirmed",
    "source_type": "unknown",
    "evidence_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "confidence": "high",
    "second_hand": false,
    "note": "No result on that evaluation was found in the official material"
   }
  ],
  "other_evaluations": [
   {
    "label": "Digital-twin protocol cases",
    "category": "case-study",
    "detail": "Three dye-and-water liquid-handling cases: catching protocol errors in advance, recovering mid-run from a fault, and diagnosing subtle physical failures afterwards",
    "source_type": "self"
   },
   {
    "label": "Wet-lab readout of the closed loop",
    "category": "wet-lab-validation",
    "detail": "Aptamer loops on two known targets and a binder experiment on one target, all designed and read out by the authors",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report that a semantic digital twin makes agent-run wet labs safer and that closed-loop design raises hit rates round on round.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "No third-party evaluation or reproduction was found; the experiments were designed and read out by the authors",
  "background": {
   "org_profile": "The developer is a large technology company's internal research division with labs in several regions; the location of this project's wet-lab work is not public.",
   "funding": "Internal company budget; no separate funding.",
   "products_partnerships": "The blog lists no outside partners; related work from the same company includes Co-Scientist and AlphaProteo.",
   "related_models": [
    "model-co-scientist",
    "model-alphaproteo",
    "model-bindcraft"
   ]
  },
  "reading_notes": [
   "This is method and infrastructure rather than a model that can be ranked, so the absence of board figures is expected.",
   "The improvement in hit rate comes with a clear error-correction narrative, but the sample is two known targets in a molecule class with extensive prior literature.",
   "The mid-loop drop on the second target shows the agent can be led astray by its own design rules and recovers only through experimental feedback.",
   "Unlike competition-style wet-lab evaluations, these experiments are designed and read out by the authors with no blind third party."
  ],
  "caveats": [
   "The binder side must separate two things: a new experiment on one target, and a computational analysis of 15 targets and 3,666 designs on existing data.",
   "The aptamer loop's intermediate round was previously omitted and all three rounds must be shown.",
   "The second target's figures pair specificity with hit rate and must not be read as three plain hit rates.",
   "It reports no benchmark on this site; the six related entries are treated as unconfirmed."
  ],
  "sources": [
   {
    "url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "title": "Official research blog (2026-10-16)",
    "kind": "blog",
    "date": "2026-10-16"
   },
   {
    "url": "https://arxiv.org/abs/2610.32755",
    "title": "Main preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2610.24806",
    "title": "Companion paper: aptamer loop",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2610.33195",
    "title": "Companion paper: binder design",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://arxiv.org/html/2610.33195v1",
    "title": "Companion paper full text (binder design)",
    "kind": "preprint",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "wetlab_validation",
    "was": "The binder entry merged new experiments with computational analysis of existing data",
    "now": "Split into two items: a new experiment on one target (5,000-member library, 1,250 designs per arm) and computational validation on existing data (15 targets, 3,666 designs)",
    "source_url": "https://arxiv.org/html/2610.33195v1",
    "date_found": "2026-11-16"
   },
   {
    "field": "wetlab_validation",
    "was": "The aptamer closed loop recorded only the first and last rounds",
    "now": "Added the middle round, and wrote the second target's numbers as specificity paired with hit rate",
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.code",
   "open_status.weights",
   "training_data.a_public",
   "release_milestones.paper_v1"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-verid/",
   "en": "/bio/benchmark/models/verid/"
  },
  "names_full_null_reason": "The name is an acronym recorded as the official pages write it, not expanded",
  "n_benchmarks_reported": {
   "counted": 0,
   "by_relation": {
    "disputed-or-unconfirmed": 6
   }
  },
  "global_entity_null_reason": "No matching page found in the site-wide entity library",
  "status_head": null
 },
 {
  "id": "model-vina",
  "kind": "model",
  "short_slug": "vina",
  "names": {
   "display": "Vina",
   "full": "AutoDock Vina",
   "full_null_reason": null,
   "aliases": [
    "AutoDock Vina",
    "Vina 1.2.x"
   ],
   "ambiguous_aliases": [
    "Vina",
    "AutoDock"
   ]
  },
  "family": null,
  "family_null_reason": "A single program; the other components of the AutoDock Suite are outside this record",
  "domain": "structure",
  "domains_secondary": [
   "small-molecule"
  ],
  "one_liner": "Classical docking program, not deep learning, and the field's de facto baseline",
  "org": {
   "names": [
    "Scripps Research"
   ],
   "global_entity_ids": []
  },
  "org_type": "academic-lab",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Classical docking and scoring function",
  "open_status": {
   "code": "open",
   "weights": "open",
   "weights_license": null,
   "code_license": "Apache-2.0 (per the repository)",
   "license_changed": null,
   "note": "Source is open; there are no deep-learning weights, and the scoring-function parameters ship with the source"
  },
  "first_release": {
   "date": "2009-06-04",
   "precision": "day",
   "event_kind": "journal-publication",
   "basis": "Online date of the original paper (the issue is dated 2010-01-30)",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1002/jcc.21334",
   "verification": "verified",
   "note": "The site said 2010, which is the issue year"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2009-06-04",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Original paper online",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1002/jcc.21334",
    "verification": "verified",
    "note": null
   },
   "paper_v1": null,
   "journal_online": {
    "date": "2009-06-04",
    "precision": "day",
    "event_kind": "journal-publication",
    "basis": "Original paper online",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1002/jcc.21334",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2021",
    "precision": "year",
    "event_kind": "new-version",
    "basis": "The year of the 1.2.0 paper; the repository release date is unverified",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1021/acs.jcim.1c00203",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-02-26",
   "precision": "day",
   "event_kind": "code-release",
   "basis": "GitHub release v1.2.7",
   "description": "v1.2.7 was published 2025-02-26 after v1.2.6 on 2025-01-13; commits through 2026-10-15 are maintenance",
   "timestamp_utc": null,
   "source_url": "https://github.com/ccsb-scripps/AutoDock-Vina/releases",
   "verification": "partial",
   "note": "The release notes were not read, so the description is partly verified"
  },
  "status": "maintained",
  "status_note": "Tagged releases, with recent commits as maintenance",
  "what_it_does": "Finds ligand poses inside a given pocket box with an empirical scoring function and iterative local search; it is fast and easy to run. In AI comparisons it is the floor baseline for physics scoring with a known pocket, set against the generative docking of DiffDock and against co-folding models.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "The scoring function was fitted on the PDBbind refined set; there is no large deep-learning training set",
    "source_url": "https://doi.org/10.1002/jcc.21334"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-05-posebusters",
    "relation_type": "cited-as-comparison-only",
    "source_type": "third",
    "counted": false,
    "eval_set": "PoseBusters V2 with known pocket, from a third-party comparison table",
    "source_url": "https://arxiv.org/abs/2412.10743",
    "confidence": "medium",
    "verification": "verified",
    "note": "The developers self-report nothing; the figures all come from other papers, and a known pocket is not comparable with blind prediction"
   },
   {
    "benchmark_id": "bench-12-dude",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "LIT-PCBA, rerun by a third-party study",
    "source_url": "https://arxiv.org/pdf/2605.01681",
    "confidence": "medium",
    "verification": "verified",
    "note": "A third-party study ran this program and a derived scoring tool on a screening task"
   },
   {
    "benchmark_id": "bench-33-posex",
    "relation_type": "retrospective-dataset-use",
    "source_type": "third",
    "counted": true,
    "eval_set": "PoseX subset, evaluated by the benchmark authors",
    "source_url": "https://arxiv.org/html/2505.01700v3",
    "confidence": "medium",
    "verification": "verified",
    "note": "A pairing added this round, evaluated on a subset by the PoseX paper"
   }
  ],
  "other_evaluations": [
   {
    "label": "Role as the traditional-method baseline",
    "category": "third-party-evaluation",
    "detail": "It recurs as the floor baseline in deep-learning docking and co-folding papers, usually with the pocket known",
    "source_type": "third"
   },
   {
    "label": "How the scoring function was fitted",
    "category": "self-report",
    "detail": "The original paper states the scoring function was fitted on a public refined complex set using more than linear regression (authors' account)",
    "source_type": "self"
   },
   {
    "label": "Suite maintenance",
    "category": "resource-use",
    "detail": "The AutoDock Suite is developed jointly by four laboratories and coordinated by the team at this record's institute",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report fast pose generation inside a given pocket from an empirical scoring function plus iterative local search, with a low barrier to use (per the original and 1.2.x papers).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "Third-party papers and benchmark authors have rerun it on several datasets, with conclusions that shift by task and pocket setting (no figures).",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "An academic project maintained as open source by the institute's docking-tool team in La Jolla, California; the AutoDock Suite is developed jointly by four laboratories under that team's coordination.",
   "funding": null,
   "funding_null_reason": "Academic project; funding amounts undisclosed",
   "products_partnerships": "Version 1.2.0 was published as a methods paper in 2021 and minor releases continue; it is the most common traditional comparator in docking and co-folding papers.",
   "products_partnerships_null_reason": null,
   "related_models": [
    "model-diffdock",
    "model-neuralplexer3",
    "model-alphafold3"
   ]
  },
  "reading_notes": [
   "Its figures assume a known pocket while co-folding models mostly predict blind; the conditions differ.",
   "Classical scoring is not penalised on physical validity; in virtual screening it depends on the task.",
   "It works as a floor baseline, not as a primary competitor.",
   "It is not deep learning, so there are no weights or training scales to compare."
  ],
  "caveats": [
   "The developers never self-report on the site's benchmarks; every relation comes from third parties.",
   "The latest release's notes were not read, so the description is only partly verified.",
   "The name is a common word and must pass the same-name guard before auto-linking."
  ],
  "sources": [
   {
    "url": "https://doi.org/10.1002/jcc.21334",
    "title": "Original docking paper",
    "kind": "journal",
    "date": {
     "date": "2009-06-04",
     "precision": "day",
     "event_kind": "journal-publication",
     "basis": "Online date",
     "timestamp_utc": null,
     "source_url": "https://doi.org/10.1002/jcc.21334",
     "verification": "verified",
     "note": null
    }
   },
   {
    "url": "https://doi.org/10.1021/acs.jcim.1c00203",
    "title": "Methods paper for version 1.2.0",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://github.com/ccsb-scripps/AutoDock-Vina/releases",
    "title": "AutoDock Vina GitHub releases",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://ccsb.scripps.edu/autodocksuite",
    "title": "Official AutoDock Suite page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://arxiv.org/abs/2412.10743",
    "title": "Third-party comparison table in a co-folding paper",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2010 (the journal issue year)",
    "now": "2009-06-04, the original paper's online date",
    "source_url": "https://doi.org/10.1002/jcc.21334",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update",
    "was": "2025-03-09",
    "now": "2025-02-26, the publication date of v1.2.7",
    "source_url": "https://github.com/ccsb-scripps/AutoDock-Vina/releases",
    "date_found": "2026-11-16"
   },
   {
    "field": "latest_update.description",
    "was": "the previous release v1.2.6 was 2025-02-15",
    "now": "the previous release v1.2.6 was 2025-01-13",
    "source_url": "https://github.com/ccsb-scripps/AutoDock-Vina/releases",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "release_milestones.code_release.date",
   "latest_update.description"
  ],
  "as_of": "2026-11-16",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-vina/",
   "en": "/bio/benchmark/models/vina/"
  },
  "names_full_null_reason": null,
  "n_benchmarks_reported": null,
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-x-cell",
  "kind": "model",
  "short_slug": "x-cell",
  "names": {
   "display": "X-Cell",
   "full": null,
   "aliases": [
    "X-Cell",
    "X-Cell-Ultra",
    "X-Cell Mini"
   ],
   "ambiguous_aliases": []
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "X-Cell-Ultra",
     "id_or_null": null,
     "first_release": {
      "date": "2026-03-17",
      "precision": "day",
      "year_hint": null,
      "basis": "Developer press release",
      "event_kind": "product-update",
      "timestamp_utc": null,
      "source_url": "https://xaira-therapeutics.github.io/X-Cell",
      "verification": "partial",
      "note": null
     }
    },
    {
     "name": "X-Cell Mini",
     "id_or_null": null,
     "first_release": {
      "date": null,
      "precision": null,
      "year_hint": null,
      "basis": "Planned, not yet released",
      "event_kind": "none-since-release",
      "timestamp_utc": null,
      "source_url": "https://xaira-therapeutics.github.io/X-Cell",
      "verification": "unverified",
      "note": null
     }
    }
   ],
   "boundary": "Two sizes from one release: the largest at about 4.9 billion parameters and a small version initialised from a public single-cell model and not yet released"
  },
  "family_null_reason": null,
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "Virtual-cell model predicting expression change after genetic perturbation",
  "org": {
   "names": [
    "Xaira Therapeutics"
   ],
   "global_entity_ids": [
    "xaira-therapeutics"
   ]
  },
  "org_type": "ai-native-biotech",
  "org_type_note": null,
  "country": [
   "US"
  ],
  "model_class": "Single-cell perturbation prediction (diffusion language model)",
  "open_status": {
   "code": "closed",
   "weights": "closed",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Weights and inference code are marked as coming soon under a planned non-commercial share-alike licence; the test perturbation set is already public"
  },
  "first_release": {
   "date": "2026-03-17",
   "precision": "day",
   "year_hint": null,
   "basis": "Developer press release, with the repository public a day earlier",
   "event_kind": "product-update",
   "timestamp_utc": null,
   "source_url": "https://xaira-therapeutics.github.io/X-Cell",
   "verification": "partial",
   "note": "Both dates rest on a company release and the repository, with no third-party corroboration"
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2026-03-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Code repository made public (two commits only)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/Xaira-Therapeutics/X-Cell",
    "verification": "partial",
    "note": "The repository date is recorded only and is not used as first-release evidence"
   },
   "paper_v1": {
    "date": "2026-03-18",
    "precision": "day",
    "year_hint": null,
    "basis": "Preprint",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.03.18.712807",
    "verification": "unverified",
    "note": null
   },
   "journal_online": null,
   "code_release": {
    "date": "2026-03-16",
    "precision": "day",
    "year_hint": null,
    "basis": "Repository made public",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/Xaira-Therapeutics/X-Cell",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": {
    "date": "2026-03-17",
    "precision": "day",
    "year_hint": null,
    "basis": "Company press release",
    "event_kind": "product-update",
    "timestamp_utc": null,
    "source_url": "https://www.biospace.com/press-releases/xaira-therapeutics-launches-x-cell-its-first-virtual-cell-model-trained-on-the-largest-ever-genome-wide-perturbation-dataset-x-atlas-pisces",
    "verification": "partial",
    "note": null
   }
  },
  "latest_update": {
   "date": "2026-03-17",
   "precision": "day",
   "year_hint": null,
   "basis": "The release event itself",
   "event_kind": "none-since-release",
   "timestamp_utc": null,
   "source_url": "https://xaira-therapeutics.github.io/X-Cell",
   "verification": "partial",
   "description": "No substantive event after the release was found as of 2026-11-16; the project page still lists weights and inference code as coming soon.",
   "note": "Same-day latest update means only that no later event was found"
  },
  "status": "active",
  "status_note": "Released within 12 months, with no event found since",
  "what_it_does": "Iteratively replaces control expression values with post-perturbation values using a diffusion language model, bringing in gene and protein interaction priors through cross-attention; the preprint reports a power law in perturbation prediction with scale and zero-shot transfer to two new cell types (second-hand). It differs from State and Tahoe-x1 in architecture and in training on data the developer produced itself.",
  "training_data": {
   "a_public": {
    "value": "not-disclosed",
    "detail": "The project page lists only the in-house perturbation set; the small version is initialised from a public single-cell model and so inherits its public data indirectly, with other public data use undisclosed",
    "source_url": "https://xaira-therapeutics.github.io/X-Cell"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "yes",
    "detail": "An in-house CRISPRi perturbation set of 25.6 million cells; the context count conflicts, with the project page and press release giving seven screens (9,735-18,924 perturbations each) and most reports giving 16 biological contexts",
    "source_url": "https://xaira-therapeutics.github.io/X-Cell"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "self-reported-modified-protocol",
    "source_type": "self",
    "evidence_url": "https://xaira-therapeutics.github.io/X-Cell",
    "confidence": "high",
    "second_hand": false,
    "note": "A self-built held-out iPSC perturbation set rather than the challenge's blind test; the project page claims a multiple over the runner-up without naming it, while the preprint abstract frames the multiple as a maximum"
   }
  ],
  "other_evaluations": [
   {
    "label": "Held-out iPSC perturbation set",
    "category": "own-task-set",
    "detail": "A held-out perturbation set built by the developer, with training and evaluation from the same source",
    "source_type": "self"
   },
   {
    "label": "Zero-shot transfer to new cell types",
    "category": "case-study",
    "detail": "Zero-shot evaluation on two unseen cell types, per second-hand reports",
    "source_type": "third"
   }
  ],
  "claimed_advantage": "Developers report: training on an in-house genome-wide perturbation dataset, a power law in perturbation prediction with model size, and a large margin over an unnamed runner-up on a held-out perturbation set.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": null,
  "independent_evidence_null_reason": "Weights are not public, no third-party evaluation was found, and it has no result on the public blind-test challenge",
  "background": {
   "org_profile": "The developer is a US AI-first drug discovery company launched in April 2024, working on protein and antibody drugs and continuing a university lab's protein design lineage; headcount is not officially disclosed, with a third-party data provider estimating about 193 in March 2026.",
   "funding": "It launched with over $1B in Series A commitments co-led by two venture firms; the valuation is undisclosed.",
   "products_partnerships": "Its data products are two generations of perturbation atlas (about 8 million cells, then 25.6 million), and its models are this one at about 4.9 billion parameters plus a planned small version; it competes in perturbation prediction with State and Tahoe-x1.",
   "related_models": [
    "model-state",
    "model-tahoe-x1",
    "model-scgpt-geneformer"
   ]
  },
  "reading_notes": [
   "The held-out result is self-reported and the project page names no competitor, so it is not evidence of a lead.",
   "The in-house dataset is a moat, but training and evaluation share a source and cross-cell-type generalisation has no independent reproduction.",
   "With no public blind-test result it cannot be compared with peers under one protocol.",
   "The context count appears as both seven and 16, so any citation must state which basis it uses."
  ],
  "caveats": [
   "Weights and inference code are unreleased, so nothing can be rechecked.",
   "Both first release and latest update rest on a company release and the repository.",
   "Sources disagree on the number of contexts in the training set.",
   "Its only pairing here is a self-built held-out set, not a challenge result.",
   "No event found since release is not the same as the project having stalled."
  ],
  "sources": [
   {
    "url": "https://xaira-therapeutics.github.io/X-Cell",
    "title": "Project page",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://www.biospace.com/press-releases/xaira-therapeutics-launches-x-cell-its-first-virtual-cell-model-trained-on-the-largest-ever-genome-wide-perturbation-dataset-x-atlas-pisces",
    "title": "Company press release",
    "kind": "press",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.64898/2026.03.18.712807",
    "title": "Preprint 2026",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/Xaira-Therapeutics/X-Cell",
    "title": "Code repository",
    "kind": "repository",
    "date": null
   }
  ],
  "corrections": [],
  "unverified_fields": [
   "first_release",
   "release_milestones.paper_v1",
   "release_milestones.code_release",
   "latest_update",
   "training_data.a_public",
   "training_data.c_own_wetlab_for_training",
   "other_evaluations"
  ],
  "as_of": "2026-11-16",
  "global_entity": "xaira-therapeutics",
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-x-cell/",
   "en": "/bio/benchmark/models/x-cell/"
  },
  "names_full_null_reason": "A product name with no expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "self-reported-modified-protocol": 1
   }
  },
  "global_entity_null_reason": null,
  "status_head": null
 },
 {
  "id": "model-xtrimopglm",
  "kind": "model",
  "short_slug": "xtrimopglm",
  "names": {
   "display": "xTrimoPGLM",
   "full": null,
   "aliases": [
    "xTrimoPGLM",
    "xTrimoPGLM-100B",
    "xT-Fold",
    "xTrimoPGLM-Fold"
   ],
   "ambiguous_aliases": [
    "xTrimo"
   ]
  },
  "family": null,
  "family_null_reason": "Several sizes and the folding branch are treated as versions within one record",
  "domain": "protein",
  "domains_secondary": [
   "structure"
  ],
  "one_liner": "A 100B unified protein language model from BioMap and partners for understanding and generation, with a folding branch",
  "org": {
   "names": [
    "BioMap",
    "Tsinghua University",
    "MBZUAI"
   ],
   "global_entity_ids": [
    "biomap"
   ]
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "CN",
   "AE"
  ],
  "model_class": "Protein LM / folding",
  "open_status": {
   "code": "partial",
   "weights": "unclear",
   "weights_license": null,
   "code_license": null,
   "license_changed": null,
   "note": "Some sizes have weights on Hugging Face and several are in the ProteinGym table; licence not retrieved this round"
  },
  "first_release": {
   "date": "2023-07-06",
   "precision": "day",
   "year_hint": null,
   "basis": "bioRxiv preprint v1",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.07.05.547496",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-07-06",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.07.05.547496",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2023-07-06",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1 (arXiv version 2024-01)",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.07.05.547496",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2025-04-03",
    "precision": "day",
    "year_hint": null,
    "basis": "Nature Methods online",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://doi.org/10.1038/s41592-025-02636-z",
    "verification": "verified",
    "note": null
   },
   "code_release": {
    "date": "2025-03",
    "precision": "month",
    "year_hint": null,
    "basis": "First GitHub commit (reference only, not used as first release)",
    "event_kind": "code-release",
    "timestamp_utc": null,
    "source_url": "https://github.com/biomap-research/xTrimoPGLM",
    "verification": "partial",
    "note": null
   },
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-04-03",
   "precision": "day",
   "year_hint": null,
   "basis": "Nature Methods online",
   "event_kind": "journal-publication",
   "timestamp_utc": null,
   "source_url": "https://doi.org/10.1038/s41592-025-02636-z",
   "verification": "verified",
   "description": "Journal version published; arXiv v2 had been posted on 2024-12-09",
   "note": null
  },
  "status": "maintained",
  "status_note": "Most recent sourced event is the 2025-04 journal paper",
  "what_it_does": "Combines masked-language and autoregressive (GLM) objectives in one 100B-parameter model for both protein representation and sequence generation; the xT-Fold branch predicts structure from language-model representations. Compared with ESM3 it takes the single very large model route, far larger than contemporaneous ESM models.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "About 939M protein sequences, 200B tokens: UniRef90 plus ColabFoldDB clusters and members",
    "source_url": "https://arxiv.org/abs/2401.06199"
   },
   "b_private_licensed": {
    "value": "no",
    "detail": "The paper lists only two public databases",
    "source_url": "https://arxiv.org/abs/2401.06199"
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2401.06199"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": "https://arxiv.org/abs/2401.06199"
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-02-proteingym",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "confidence": "high",
    "second_hand": false,
    "note": "Several sizes appear in the ProteinGym zero-shot substitution summary table; the 100B size does not beat the group's smaller ones and the CLM branch is much lower"
   },
   {
    "benchmark_id": "bench-26-flip",
    "relation_type": "self-reported-official-protocol",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2401.06199",
    "confidence": "high",
    "second_hand": false,
    "note": "Only the FLIP GB1 sampled split, after fine-tuning; not zero-shot and not the full FLIP suite"
   },
   {
    "benchmark_id": "bench-01-casp",
    "relation_type": "retrospective-dataset-use",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2401.06199",
    "confidence": "high",
    "second_hand": false,
    "note": "Folding branch evaluated post hoc on 56 CASP15 targets; not participation"
   },
   {
    "benchmark_id": "bench-15-cameo",
    "relation_type": "retrospective-dataset-use",
    "source_type": "self",
    "evidence_url": "https://arxiv.org/abs/2401.06199",
    "confidence": "high",
    "second_hand": false,
    "note": "Folding branch evaluated post hoc on a static set of 194 CAMEO targets from 2022-04 to 06"
   }
  ],
  "other_evaluations": [
   {
    "label": "18 protein understanding tasks",
    "category": "own-task-set",
    "detail": "The authors report beating prior state of the art on 15 of 18; mostly fine-tuned, not comparable with zero-shot boards",
    "source_type": "self"
   },
   {
    "label": "Folding branch vs single-sequence methods",
    "category": "comparison-vs-baseline-only",
    "detail": "Compared with ESMFold and OmegaFold; still behind MSA/template methods (author-reported)",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: one 100B model covers understanding and generation, beats prior state of the art on most understanding tasks, and its folding branch beats other single-sequence methods.",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The ProteinGym maintainers' summary table lists several sizes and shows the 100B size no better than smaller ones, consistent with the observation that pLM scaling plateaus after roughly 1B to 4B.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "BioMap was founded in Beijing in 2020 as an AI life-science company that positions itself as an enabler rather than a drug developer; paper authors also come from Tsinghua University and MBZUAI.",
   "funding": "A $100M Series A in 2021 led by GGV; a 2024-06 strategic round of undisclosed size; a 2026-03 media report of a confidential Hong Kong IPO filing, not announced by the company.",
   "products_partnerships": "A 2023-10 Sanofi deal for AI modules ($10M upfront, milestones over $1B); the series also includes xTrimoSCPerturb/scFoundation, and a multimodal xTrimo V3 was announced in 2026-10.",
   "related_models": [
    "model-esm3",
    "model-xtrimoscperturb-scfoundation"
   ]
  },
  "reading_notes": [
   "The 100B scale shows no edge on ProteinGym; in the official table it trails the group's own 3B size.",
   "Folding results are small-sample, self-reported post-hoc evaluations, not CASP/CAMEO entries.",
   "FLIP covers only the GB1 subtask after fine-tuning and does not mean leading on the full suite."
  ],
  "caveats": [
   "\"Ahead on 15 of 18\" is author-reported and mostly fine-tuned.",
   "CASP15 and CAMEO are post-hoc data reuse, not counted as participation.",
   "Weight availability and licences were not checked item by item.",
   "An earlier site figure of \"about 94B unique sequences\" was an order-of-magnitude error and is corrected from the paper."
  ],
  "sources": [
   {
    "url": "https://arxiv.org/abs/2401.06199",
    "title": "arXiv preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41592-025-02636-z",
    "title": "Nature Methods paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.07.05.547496",
    "title": "bioRxiv version record",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/OATML-Markslab/ProteinGym/blob/main/benchmarks/DMS_zero_shot/substitutions/Spearman/Summary_performance_DMS_substitutions_Spearman.csv",
    "title": "ProteinGym official summary table",
    "kind": "leaderboard",
    "date": null
   },
   {
    "url": "https://www.businesswire.com/news/home/20231010485897/en",
    "title": "Sanofi collaboration press release",
    "kind": "press",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "first_release",
    "was": "2024-01-11（arXiv v1）",
    "now": "2023-07-06（bioRxiv v1）",
    "source_url": "https://api.biorxiv.org/details/biorxiv/10.1101/2023.07.05.547496",
    "date_found": "2026-11-16"
   },
   {
    "field": "training_data.a_public",
    "was": "About 94B unique sequences",
    "now": "About 939 million sequences, 200 billion tokens",
    "source_url": "https://arxiv.org/abs/2401.06199",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "open_status.weights",
   "open_status.code_license",
   "open_status.weights_license",
   "release_milestones.code_release"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-xtrimopglm/",
   "en": "/bio/benchmark/models/xtrimopglm/"
  },
  "names_full_null_reason": "The paper gives no expanded name",
  "n_benchmarks_reported": {
   "counted": 2,
   "by_relation": {
    "official-participation": 1,
    "self-reported-official-protocol": 1,
    "retrospective-dataset-use": 2
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developer but none for the model itself",
  "status_head": null
 },
 {
  "id": "model-xtrimoscperturb-scfoundation",
  "kind": "model",
  "short_slug": "xtrimoscperturb-scfoundation",
  "names": {
   "display": "xTrimoSCPerturb / scFoundation",
   "full": null,
   "aliases": [
    "xTrimoSCPerturb",
    "scFoundation",
    "xTrimoscFoundationα",
    "BM_xTVC"
   ],
   "ambiguous_aliases": [
    "xTrimo"
   ]
  },
  "family": {
   "is_family": true,
   "members": [
    {
     "name": "scFoundation",
     "id_or_null": null,
     "first_release": {
      "date": "2023-05-31",
      "precision": "day",
      "year_hint": null,
      "basis": "bioRxiv preprint v1",
      "event_kind": "paper-revision",
      "timestamp_utc": null,
      "source_url": "https://www.biorxiv.org/content/10.1101/2023.05.29.542705",
      "verification": "verified",
      "note": null
     }
    },
    {
     "name": "xTrimoSCPerturb",
     "id_or_null": null,
     "first_release": {
      "date": "2025-12",
      "precision": "month",
      "year_hint": null,
      "basis": "Ranked in the official VCC 2025 wrap-up (no standalone release)",
      "event_kind": "results-published",
      "timestamp_utc": null,
      "source_url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up",
      "verification": "partial",
      "note": null
     }
    }
   ],
   "boundary": "scFoundation is the public foundation model; xTrimoSCPerturb is an unreleased challenge solution built on a modified version"
  },
  "family_null_reason": null,
  "domain": "single-cell",
  "domains_secondary": [],
  "one_liner": "BioMap's single-cell foundation model scFoundation and the VCC 2025-winning perturbation entry built on it",
  "org": {
   "names": [
    "BioMap",
    "Tsinghua University"
   ],
   "global_entity_ids": [
    "biomap"
   ]
  },
  "org_type": "mixed",
  "org_type_note": null,
  "country": [
   "CN"
  ],
  "model_class": "Single-cell FM / perturbation",
  "open_status": {
   "code": "partial",
   "weights": "open",
   "weights_license": "Apache-2.0",
   "code_license": "Apache-2.0",
   "license_changed": null,
   "note": "scFoundation code and weights are public (Apache-2.0); xTrimoSCPerturb has no public repository or weights (checked 2026-11-16)"
  },
  "first_release": {
   "date": "2023-05-31",
   "precision": "day",
   "year_hint": null,
   "basis": "scFoundation bioRxiv preprint v1",
   "event_kind": "paper-revision",
   "timestamp_utc": null,
   "source_url": "https://www.biorxiv.org/content/10.1101/2023.05.29.542705",
   "verification": "verified",
   "note": null
  },
  "release_milestones": {
   "first_public_evidence": {
    "date": "2023-05-31",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2023.05.29.542705",
    "verification": "verified",
    "note": null
   },
   "paper_v1": {
    "date": "2023-05-31",
    "precision": "day",
    "year_hint": null,
    "basis": "bioRxiv preprint v1",
    "event_kind": "paper-revision",
    "timestamp_utc": null,
    "source_url": "https://www.biorxiv.org/content/10.1101/2023.05.29.542705",
    "verification": "verified",
    "note": null
   },
   "journal_online": {
    "date": "2024-06-06",
    "precision": "day",
    "year_hint": null,
    "basis": "Nature Methods online (registry record; issue dated 2024-08)",
    "event_kind": "journal-publication",
    "timestamp_utc": null,
    "source_url": "https://api.crossref.org/works/10.1038/s41592-024-02305-7",
    "verification": "verified",
    "note": null
   },
   "code_release": null,
   "weights_release": null,
   "challenge_open": null,
   "official_announcement": null
  },
  "latest_update": {
   "date": "2025-12",
   "precision": "month",
   "year_hint": null,
   "basis": "Official VCC 2025 wrap-up names xTrimoSCPerturb overall first",
   "event_kind": "results-published",
   "timestamp_utc": null,
   "source_url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up",
   "verification": "partial",
   "description": "The organiser published the ranking but no scores, paper or weights; the scFoundation repository had only README edits in 2025-11",
   "note": "The wrap-up's exact day was not retrieved; month-level evidence only"
  },
  "status": "active",
  "status_note": "Challenge result published within 12 months",
  "what_it_does": "scFoundation is an asymmetric Transformer of about 100M parameters pretrained on expression profiles over about 20k genes; the paper reports downstream tasks such as expression enhancement, drug response, perturbation prediction and cell-type annotation. xTrimoSCPerturb is the team's entry for the Arc Virtual Cell Challenge 2025, described as deep learning plus classical statistical features on a re-pretrained modified scFoundation; unlike general baselines such as scGPT/Geneformer it is a task-specific solution for one challenge.",
  "training_data": {
   "a_public": {
    "value": "yes",
    "detail": "scFoundation pretraining used over 50 million human single-cell transcriptomes",
    "source_url": "https://doi.org/10.1038/s41592-024-02305-7"
   },
   "b_private_licensed": {
    "value": "not-disclosed",
    "detail": null,
    "source_url": null
   },
   "c_own_wetlab_for_training": {
    "value": "no",
    "detail": "The H1 perturbation data were produced by Arc and issued to teams for fine-tuning, not generated by the developer",
    "source_url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up"
   },
   "d_wetlab_validation_only": {
    "value": "no",
    "detail": null,
    "source_url": null
   }
  },
  "wetlab_validation": [],
  "benchmarks_reported": [
   {
    "benchmark_id": "bench-24-vcc",
    "relation_type": "official-participation",
    "source_type": "board",
    "evidence_url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up",
    "confidence": "high",
    "second_hand": false,
    "note": "Ranked overall first in the organiser's official VCC 2025 wrap-up, with no per-metric values published. A separate independent study compared scFoundation with simple linear baselines, with unfavourable results"
   }
  ],
  "other_evaluations": [
   {
    "label": "Independent perturbation-prediction study",
    "category": "third-party-evaluation",
    "detail": "Five foundation models plus two deep models against deliberately simple linear baselines on single and double perturbations; none beat the baseline, scFoundation included",
    "source_type": "third"
   },
   {
    "label": "scFoundation paper downstream tasks",
    "category": "own-task-set",
    "detail": "Expression enhancement, drug response, cell-type annotation, gene-module inference and others; author-reported",
    "source_type": "self"
   }
  ],
  "claimed_advantage": "Developers report: scFoundation leads on many single-cell downstream tasks; xTrimoSCPerturb ranked first overall in VCC 2025 (the latter published by the organiser).",
  "claimed_advantage_null_reason": null,
  "independent_evidence": "The organiser confirmed the VCC 2025 ranking but its wrap-up notes pure AI methods did not consistently beat statistical baselines; an independent study found scFoundation did not beat simple linear baselines.",
  "independent_evidence_null_reason": null,
  "background": {
   "org_profile": "BioMap was founded in 2020, is headquartered in Beijing with R&D sites in Suzhou, Hong Kong and Silicon Valley, and builds the xTrimo life-science foundation models; scFoundation also has Tsinghua University authors.",
   "funding": "A Series A above $100M in 2021; a 2024 strategic partnership and a further round reported with undisclosed amounts; a 2026-03 media report of a confidential Hong Kong IPO filing, not announced by the company.",
   "products_partnerships": "A 2023-10 Sanofi deal for biologics AI modules ($10M upfront, potential milestones over $1B); the protein language model xTrimoPGLM is a sibling.",
   "related_models": [
    "model-xtrimopglm",
    "model-scgpt-geneformer",
    "model-state"
   ]
  },
  "reading_notes": [
   "The VCC first place is a genuine official result, but the organiser itself notes pure AI methods did not consistently beat statistical baselines.",
   "With no public scores or weights, the rank cannot be turned into comparable accuracy and cannot be reproduced.",
   "Independent evaluations of scFoundation are generally unfavourable, a problem shared across single-cell foundation models."
  ],
  "caveats": [
   "xTrimoSCPerturb is unreleased with no paper; architecture and scores are not public.",
   "The site's Nature Methods 2024-08 is the issue month; online publication was 2024-06-06.",
   "Training data are observational only, with no self-generated perturbation set."
  ],
  "sources": [
   {
    "url": "https://arcinstitute.org/news/virtual-cell-challenge-2025-wrap-up",
    "title": "Official VCC 2025 wrap-up",
    "kind": "official-site",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41592-024-02305-7",
    "title": "scFoundation journal paper",
    "kind": "journal",
    "date": null
   },
   {
    "url": "https://www.biorxiv.org/content/10.1101/2023.05.29.542705",
    "title": "scFoundation preprint",
    "kind": "preprint",
    "date": null
   },
   {
    "url": "https://github.com/biomap-research/scFoundation",
    "title": "scFoundation repository",
    "kind": "repository",
    "date": null
   },
   {
    "url": "https://doi.org/10.1038/s41592-025-02772-6",
    "title": "Independent perturbation-prediction evaluation paper",
    "kind": "third-party-paper",
    "date": null
   }
  ],
  "corrections": [
   {
    "field": "release_milestones.journal_online",
    "was": "2024-08 (volume issue)",
    "now": "2024-06-06 (published online)",
    "source_url": "https://api.crossref.org/works/10.1038/s41592-024-02305-7",
    "date_found": "2026-11-16"
   }
  ],
  "unverified_fields": [
   "latest_update",
   "open_status.weights_license",
   "training_data.b_private_licensed",
   "background.funding"
  ],
  "as_of": "2026-11-17",
  "global_entity": null,
  "site_urls": {
   "zh": "/AI4S/bio/benchmarks/model-xtrimoscperturb-scfoundation/",
   "en": "/bio/benchmark/models/xtrimoscperturb-scfoundation/"
  },
  "names_full_null_reason": "Neither name has an expanded form",
  "n_benchmarks_reported": {
   "counted": 1,
   "by_relation": {
    "official-participation": 1
   }
  },
  "global_entity_null_reason": "The site-wide entity library has a page for the developer but none for the model itself",
  "status_head": null
 }
]
