AI for Bio · Models

AI for Bio benchmarks: all 112 models

All 112 models in the survey, grouped by domain. “Benchmarks” counts established pairings with the 44 benchmarks.

Structure & co-folding (24)

ModelOrganizationCodeTraining dataBenchmarks
AF2-Multimer
Multi-chain fine-tune of AF2; the classic protein-complex baseline
Google DeepMindOpenPublic data only1
AlphaFold 3
Biomolecular co-folding model that later co-folding models use as the reference baseline
Google DeepMind, Isomorphic LabsPartly openPublic data only6
BioEmu-2
All-atom diffusion model sampling structural ensembles of complexes and nucleic acids
Microsoft Research AI for ScienceOpenPublic data only0
Boltz-1 / Boltz-1x
First fully open co-folding model approaching AF3, with 1x adding physical-validity steering
MIT, Boltz PBCOpenPublic data only2
Boltz-2
Open model that adds a small-molecule binding-affinity head on top of co-folding
MIT, Recursion, Boltz PBCOpenPublic data only3
Chai-1
Open-weights co-folding model that stays stable without MSAs and accepts experimental restraints
Chai DiscoveryOpenPublic data only5
Chai-2
Closed zero-shot antibody and binder design model with an upgraded folding module built in
Chai DiscoveryClosedPublic data + wet lab for validation only0
DiffDock
Diffusion-based blind docking, now used mainly as the deep-learning docking baseline
MIT CSAILOpenPublic data only2
ESMFold2
Protein-language-model co-folding that needs no MSA retrieval, released alongside a sequence atlas
Biohub, EvolutionaryScale teamUnclearPublic data + wet lab for validation only2
HelixFold-S1
A contact-guided sampling method added on top of HelixFold3, not a new base model
BaiduClosedPublic data only0
IntelliFold-2
Hong Kong co-folding model billed as an AF3 improvement; Pro tier is server-only
IntelliGen AIOpenPublic data only2
IsoDDE
Isomorphic Labs closed drug design engine extending AlphaFold 3 co-folding and affinity
Isomorphic LabsClosedNot disclosed2
MassiveFold
Large-scale sampling and parallelisation around AF2, not a new model; hard interfaces gain most
Lille University UGSF, CNRSOpenPublic data only2
NeuralPLexer3
Flow-matching co-folding model emphasising stereochemistry and physical validity
Iambic TherapeuticsClosedPublic data only3
OpenComplex2
All-atom generative model that samples conformational distributions of proteins, nucleic acids and complexes by graph diffusion rather than predicting one structure
Beijing Academy of Artificial IntelligenceUnclearNot disclosed1
OpenDDE
Fully open 655M-parameter co-folding model tuned for antibody-antigen, billed as an open discovery engine
Aureka AI Research, Aureka BiotechnologiesOpenPublic data only1
OpenFold3
Consortium-built Apache-2.0 open reproduction of AlphaFold 3 with a federated private-data fine-tuning track
OpenFold Consortium, Open Molecular Software Foundation, Columbia University, Lawrence Livermore National Laboratory, Seoul National UniversityOpenPrivate or licensed data3
Pearl
Genesis Molecular AI's closed protein-ligand co-folding model, trained on large-scale physics-based synthetic structures
Genesis Molecular AIClosedPublic data + wet lab for validation only2
Protenix
Open co-folding series reproducing and trying to exceed the contemporary closed model, with v2 weights turned proprietary
ByteDance Seed, ByteDance Seed AI for SciencePartly openPublic data + wet lab for validation only4
RF3
Open co-folding model from a protein design institute, released under BSD with the AtomWorks data framework
University of Washington Institute for Protein Design, RosettaCommonsOpenPublic data only1
RhoFold+/DRfold2/trRosettaRNA
Three independent RNA 3D prediction lines from three academic groups, not mutually comparable
CUHK / Harvard / MIT collaboration, National University of Singapore group, Shandong University (Qingdao) groupPartly openPublic data only2
RNAPro
Open model predicting all-atom RNA 3D structure, about 488M parameters
NVIDIA, Stanford UniversityOpenPublic data only1
SeedFold
ByteDance Seed AF3-style co-folding model that scales by widening the Pairformer, with a linear triangle-attention variant
ByteDance SeedUnclearPublic data only1
Vina
Classical docking program, not deep learning, and the field's de facto baseline
Scripps ResearchOpenPublic data only2

Small molecules (21)

ModelOrganizationCodeTraining dataBenchmarks
ADMET-AI
ADMET prediction platform returning 41 endpoints per molecule against approved-drug distributions
Stanford University (computer science), Greenstone BiosciencesOpenPublic data only1
AQAffinity
Open protein-ligand affinity model adding an affinity head to an open co-folding backbone
SandboxAQOpenPublic data only1
Beacon (Inductive Bio)
Inductive Bio's closed ADMET/PK predictor trained on a pre-competitive pharma data consortium
Inductive BioClosedPrivate or licensed data0
Chemformer
Pretrained molecule-string sequence model for reaction prediction and retrosynthesis, long used as a baseline
AstraZeneca (Molecular AI, Discovery Sciences)UnclearPrivate or licensed data1
Chemprop v2
Open-source molecular property prediction library on directed message passing, the most common deep-learning baseline
MIT (chemical engineering and electrical engineering groups)OpenPublic data only0
Enchant v1/v2
Closed multimodal transformer predicting molecular properties and clinically relevant endpoints from early data
Iambic TherapeuticsClosedOwn wet-lab training data0
ether0
A 24B-parameter chemistry reasoning model whose answers are molecules
FutureHouseOpenPublic data only1
KERMT
Graph-transformer chemical pretrained model with multitask fine-tuning and acceleration for in-house ADMET prediction
Merck & Co. (modeling and informatics), NVIDIA (BioNeMo team)OpenPrivate or licensed data0
MiniMol
Ten-million-parameter molecular foundation model densely pretrained on thousands of supervised tasks
Valence Labs (Recursion Pharmaceuticals)OpenPublic data only1
MolE
Molecular-graph foundation model adapting disentangled attention to graphs for property prediction
RecursionOpenPublic data only1
MoLFormer-XL
Chemical language model pretraining on billion-scale molecule strings, then fine-tuned for property prediction
IBM ResearchOpenPublic data only1
Nesso-1
Open protein-ligand affinity model trading coarse-grained structure for inference speed
Valence Labs, Recursion PharmaceuticalsUnclearPublic data only4
ReaSyn v2
Synthesisability projection model that builds a synthesis route from purchasable building blocks
NVIDIA, KAISTOpenPublic data only1
REINVENT 4
Open-source small-molecule generative design framework driven by reinforcement, transfer and curriculum learning
AstraZeneca Molecular AI (Discovery Sciences, R&D)OpenPublic data only0
RetroDFM-R
Reasoning-style retrosynthesis LLM trained with verifiable chemistry rewards, emitting precursors with a reasoning trace
Shanghai Jiao Tong University (X-LANCE Lab; School of Chemistry and Chemical Engineering), Suzhou LaboratoryOpenPublic data only1
Saturn
Generative molecular design model aimed at sample-efficient search under a limited oracle budget
EPFL (Laboratory of Artificial Chemical Intelligence)OpenPublic data only0
TamGen
Target-aware molecule generator that conditions on a protein pocket and refines compounds
Microsoft Research AI for Science, Global Health Drug Discovery Institute (GHDDI)OpenPublic data + wet lab for validation only1
TerraBind
Protein-ligand structure and affinity foundation model trading diffusion for speed
Terray TherapeuticsClosedPublic data + wet lab for validation only2
TxGemma
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
Google DeepMind, Google ResearchPartly openPublic data only1
Uni-Mol2
3D-conformation molecular pretraining model scaled to the billion-parameter range to study scaling
DP Technology, Peking University, AI for Science Institute, BeijingOpenPublic data only1
Uni-Mol3
Multi-molecular reaction foundation model covering products, retrosynthesis, conditions and yield
DP Technology, Peking UniversityOpenPrivate or licensed data1

Proteins (26)

ModelOrganizationCodeTraining dataBenchmarks
AlphaProteo
Closed de novo protein binder design system; seven targets reported successful and an eighth failed
Google DeepMind, Francis Crick InstituteClosedPublic data + wet lab for validation only0
AMPLIFY
Small protein language models arguing data quality matters more than scale
Mila, Polytechnique Montréal, AmgenOpenPublic data only0
BindCraft
Open binder design pipeline that backpropagates through a co-folding network, with no extra training
EPFL Laboratory of Protein Design and Immunoengineering, MITOpenPublic data + wet lab for validation only1
BoltzGen
Open general-purpose binder design model built on a co-folding architecture
MIT Jameel Clinic, Boltz PBCOpenPublic data + wet lab for validation only0
Chai-3
Chai Discovery's closed next-generation molecular design model for antibodies and multispecifics
Chai DiscoveryClosedPrivate or licensed data0
Claude protein-design study (general model + public tools)
A general language model orchestrates public tools to design binders, tested by outside labs
Anthropic, Adaptyv Bio, Twist BioscienceClosedPublic data + wet lab for validation only1
DPLM-2
Multimodal discrete-diffusion protein language model generating matched sequence and structure
ByteDance Research, Nanjing UniversityOpenPublic data only1
Dyno Psi-1 / Psi-Phi
Open-weight binder design model of about 200M parameters with a companion filter and agent suite
Dyno TherapeuticsOpenPublic data + wet lab for validation only0
ESM C
Sequence-only protein language model positioned as a replacement for ESM-2
Evolutionary Scale, Chan Zuckerberg BiohubOpenPublic data + wet lab for validation only1
ESM3
Multimodal generative protein language model; sequence, structure and function prompt each other
Evolutionary Scale, Chan Zuckerberg BiohubOpenPublic data + wet lab for validation only4
EVOLVEpro
Few-shot active-learning wrapper for protein engineering, about a dozen variants per round
Brigham and Women's Hospital / Harvard Medical School, Mass General Brigham Gene and Cell Therapy Institute, Beth Israel Deaconess Medical Center, MITOpenOwn wet-lab training data0
La-Proteina
All-atom protein generation with flow matching in a partially latent space, co-designable up to 800 residues
NVIDIAOpenPublic data only1
Latent-X / Latent-X2
Closed platform models for binder, macrocycle and antibody design, testing only tens of designs per target
Latent LabsClosedPublic data + wet lab for validation only0
LucaOne
Unified pretrained foundation model for DNA, RNA and protein sequences, about 1.8B parameters
Alibaba Cloud, Sun Yat-sen UniversityOpenPublic data only0
Mosaic
Open multi-objective protein design library optimising binders through combined scorers
Escalante BioOpenPublic data + wet lab for validation only1
Pinal
Natural-language-to-protein design: turns a function description into structure tokens, then a sequence
Westlake UniversityOpenPublic data + wet lab for validation only0
PoET-2
182M-parameter multimodal retrieval-augmented protein foundation model using family context
OpenProtein.AIPartly openPublic data only1
Profluent-E1
Retrieval-augmented protein encoder for mutation-effect scoring and representation
ProfluentOpenPublic data only3
ProGen3
Sparse mixture-of-experts autoregressive protein language model focused on scaling laws and alignment with laboratory data
ProfluentPartly openOwn wet-lab training data1
ProSST
Protein language model quantizing local structure into tokens with disentangled attention
Shanghai Jiao Tong University, Shanghai AI LaboratoryOpenPublic data only2
Proteina-Complexa
All-atom binder design model combining generative pretraining with test-time compute
NVIDIA (GenAIR)OpenPublic data + wet lab for validation only1
RFdiffusion2 / RFdiffusion3
Atomic-level diffusion design family: RFD2 scaffolds enzyme active sites, RFD3 designs all-atom interfaces
University of Washington Institute for Protein Design, Rosetta CommonsOpenPublic data + wet lab for validation only1
SaProt
Structure-aware protein language model pairing structure tokens with amino acids
Westlake University, Westlake Representation Learning LabOpenPublic data only3
SeedProteo
Diffusion-based all-atom de novo protein design model for binders
ByteDance SeedClosedPublic data + wet lab for validation only0
VenusREM
Retrieval-augmented zero-shot mutation-effect model fusing sequence, structure and homologs
Shanghai Jiao Tong University, East China University of Science and Technology, Shanghai AI LaboratoryOpenPublic data + wet lab for validation only1
xTrimoPGLM
A 100B unified protein language model from BioMap and partners for understanding and generation, with a folding branch
BioMap, Tsinghua University, MBZUAIPartly openPublic data only4

Antibodies (7)

ModelOrganizationCodeTraining dataBenchmarks
Germinal
Open-source de novo antibody and nanobody design pipeline that redesigns only the CDRs
Stanford University (chemical engineering), Arc InstituteOpenPublic data + wet lab for validation only0
HCAbLM
Nona Biosciences language model trained only on fully human heavy-chain-only antibodies, aimed at developability
Nona Biosciences, HBM Holdings (Harbour BioMed)UnclearOwn wet-lab training data0
IgGM / IgGM-FM
Generative model for antibodies and nanobodies covering structure prediction, de novo design, maturation and humanization
Tencent AI for Life Sciences LabOpenPublic data + wet lab for validation only1
JAM / JAM-2
Closed de novo antibody and VHH design models aimed at hard multi-pass membrane targets such as GPCRs
Nabla BioClosedPublic data + wet lab for validation only0
Origin-1
Generative platform designing full-length antibodies de novo against zero-prior epitopes, combining diffusion generation, paired CDR sequence design and co-folding scoring
AbsciPartly openPublic data + wet lab for validation only0
RFantibody
De novo antibody design pipeline for specified epitopes: generate backbones, design sequences, filter by structure
University of Washington Institute for Protein DesignOpenPublic data + wet lab for validation only0
tFold
Antibody structure and design system: MSA-free structure prediction plus epitope-specific de novo design
Tencent AI for Life Sciences LabPartly openPublic data + wet lab for validation only0

Genomics & RNA (9)

ModelOrganizationCodeTraining dataBenchmarks
AlphaGenome
Sequence-to-function model reading 1 Mb of DNA for base-resolution regulatory tracks and variant effects
Google DeepMindOpenPublic data only3
AlphaMissense
Structure-augmented missense pathogenicity model with a large public variant catalogue
Google DeepMindOpenPublic data only2
BioReason
Multimodal model feeding a DNA foundation model into an LLM for pathway reasoning and variant effects
University of Toronto, Vector Institute, University Health NetworkOpenPublic data only1
Borzoi
Predicts RNA-seq coverage from DNA and scores variant effects on transcription, splicing, polyadenylation
Calico Life SciencesOpenPublic data only2
BulkFormer
Foundation model for human bulk transcriptomes at about 150 million parameters
Peking University (biomedical informatics), Jilin University, Wuhan Sports UniversityOpenPublic data only0
Evo 2
Open DNA language model with million-token context for zero-shot variant scoring and sequence generation
Arc Institute, NVIDIA, Stanford University, University of California, Berkeley, University of California, San FranciscoOpenPublic data + wet lab for validation only2
NTv3
Genome foundation model with 1 Mb context doing track prediction, gene annotation and sequence generation in one
InstaDeep (subsidiary of BioNTech)OpenPublic data + wet lab for validation only1
popEVE
Proteome-wide missense pathogenicity model with scores comparable across genes
Harvard Medical School (Blavatnik Institute), Centre for Genomic Regulation (CRG), BarcelonaOpenPublic data only1
RIBOSPAN
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
Shanghai Innovation Institute, CAS Center for Excellence in Molecular Cell Science, Shanghai Jiao Tong UniversityOpenPublic data only1

Single-cell & perturbation (12)

ModelOrganizationCodeTraining dataBenchmarks
AIDO Cell
Multi-scale, stateful virtual-cell world model, currently with two prototype cell types
GenBio AIClosedNot disclosed0
C2S-Scale
Single-cell model that turns ranked expression into "cell sentences" and fine-tunes general LLMs from 410M to 27B parameters
Yale University, Google Research, Google DeepMindOpenPublic data + wet lab for validation only1
CellFM
Human-only single-cell foundation model with 800 million parameters
Sun Yat-sen University (computer science), Chongqing University, Huawei, Singleron BiotechnologiesOpenPublic data only1
Lingshu-Cell
Alibaba DAMO Academy generative cell world model using masked discrete diffusion over transcriptomes
Alibaba DAMO AcademyClosedPublic data only1
scGPT / Geneformer
First-generation single-cell foundation-model baselines for annotation, perturbation and integration
University of Toronto, University Health Network, Vector Institute, Gladstone Institutes, UCSFOpenPublic data + wet lab for validation only2
scPRINT-2
Institut Pasteur's second-generation single-cell foundation model pretrained on 350M cells from 16 species
Institut Pasteur, CNRSOpenPublic data only0
Stack
Single-cell foundation model doing in-context learning from example cells in the prompt, without fine-tuning
Arc InstituteOpenPublic data only1
STATE
Virtual cell model predicting transcriptional response to drug or genetic perturbation across cell lines
Arc InstituteOpenPublic data only1
Tahoe-x1
Single-cell foundation model: masked-pretraining encoder trained at scale including perturbation data
Tahoe TherapeuticsOpenOwn wet-lab training data1
TranscriptFormer
CZI/Biohub cross-species generative single-cell model trained on about 110M cells from 12 species
Chan Zuckerberg Initiative, Chan Zuckerberg BiohubOpenPublic data only0
X-Cell
Virtual-cell model predicting expression change after genetic perturbation
Xaira TherapeuticsClosedOwn wet-lab training data1
xTrimoSCPerturb / scFoundation
BioMap's single-cell foundation model scFoundation and the VCC 2025-winning perturbation entry built on it
BioMap, Tsinghua UniversityPartly openPublic data only1

Molecule generation (1)

ModelOrganizationCodeTraining dataBenchmarks
GenMol
NVIDIA's masked discrete-diffusion molecule generator covering several generation tasks in one model
NVIDIAOpenPublic data only1

Text, LLMs & agents (12)

ModelOrganizationCodeTraining dataBenchmarks
Biomni
General biomedical agent: an LLM plus a tool library that plans and runs bioinformatics analyses
Stanford University, PhyloOpenPublic data + wet lab for validation only3
Co-Scientist
Google DeepMind's Gemini-based multi-agent system that generates and ranks research hypotheses
Google DeepMind, Google ResearchClosedPublic data + wet lab for validation only0
Frontier general-purpose LLMs
Category entry for general frontier LLMs (Claude, GPT, Gemini) tested on bio benchmarks
Anthropic, OpenAI, Google DeepMindClosedNot disclosed5
GPT-Rosalind
Life-sciences reasoning model offered through controlled access
OpenAIClosedNot disclosed1
Intern-S
Open multimodal scientific foundation series layering scientific corpora and specialised tokenisers on general LLMs, including a trillion-parameter mixture-of-experts version
Shanghai AI LaboratoryOpenPublic data only0
Kosmos / Edison Analysis
Edison Scientific's "AI scientist" that loops literature search and data analysis into cited reports
Edison Scientific, FutureHouseClosedPublic data only1
MedGemma / Med-Gemini
Google medical language-model family: closed Med-Gemini and open-weight MedGemma
Google Research, Google DeepMindPartly openPrivate or licensed data1
Quine
Microsoft Research's biology world model plus research harness, research use only
Microsoft ResearchClosedPublic data + wet lab for validation only0
Robin
FutureHouse multi-agent discovery system linking literature review, candidate proposal and data analysis
FutureHouseOpenPublic data + wet lab for validation only0
SciReasoner
Multimodal science reasoning model encoding 3D structure as shared tokens
Shanghai AI Laboratory, The Chinese University of Hong Kong, Shanghai Institute of Materia Medica, Chinese Academy of Sciences, Stanford University, University of OxfordUnclearPublic data only1
TxAgent
Therapeutic-reasoning agent calling 211 drug and clinical tools via ToolUniverse
Harvard Medical SchoolOpenPublic data only0
VERID
Semantic digital twin of a lab workstation that loops agents between virtual dry runs, safety checks and real wet-lab experiments
Google ResearchUnclearOwn wet-lab training data0