Overview
| Name | Biomni · aliases: Biomni, Biomni Lab, Biomni-R0 |
|---|---|
| Full name | No separate expanded name |
| Ambiguous names (not auto-linked) | None |
| Family | Not a family entry. Single-model entry, not a family |
| Organization | Stanford University, Phylo |
| Organization type | Mixed · Built in an academic lab; commercialized by the startup Phylo |
| Country | United States |
| Model class | Biomedical agent |
| Domain | Text, LLMs & agents |
| First release | · preprint Basis: bioRxiv preprint v1 (Crossref issued 2025-06-02) Note: The open-source repository may predate the preprint; unverified source |
| Other milestones | |
| Latest update | · journal Basis: Science journal publication Published in Science; Biomni Lab launched 2026-02-03 and GitHub v0.0.8 came out 2025-10-27 source The most recent sourced event, not a claim that nothing changed since. |
| Status | Active · Journal publication and product launch within 12 months |
| Openness | Code: Open Weights: Unclear Open-source platform; Biomni Lab is a commercial research preview |
| Benchmarks reported | 3 counted (modified protocol 2, retrospective 1) |
| Wet-lab validation | 1 recorded campaign |
| Unverified fields | None |
| As of | 2026-11-16 |
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.
Benchmarks reported
Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 3 of 3 pairings are counted.
| Benchmark | Relation | Eval set / protocol | Source type | Note and evidence |
|---|---|---|---|---|
| 18 LAB-Bench / LABBench2 | Developer report, modified protocol | LAB-Bench DbQA/SeqQA subsets | Self-reported | Original DbQA/SeqQA subsets with 45 dev / 315 hidden test questions and 3 runs. †unverified · evidence |
| 29 BixBench | Developer report, modified protocol | BixBench full + verified-50 | Self-reported | Run by its developer on the full set and a verified subset with its own agent harness †partly verified · evidence |
| 41 BioML-bench | Retrospective dataset use | — | Third-party | Re-run by the benchmark authors rather than reported by the developer; the Chinese matrix marks it self-reported, which under this round's definition should be third-party and is recorded as an open inconsistency. †partly verified · evidence |
Other evaluations (outside the 44)
None recorded.
Developers report: Developers report near-expert performance on several biomedical tasks and zero-shot handling of new tasks
Independent evidence: In a BioML-bench re-run by its authors, Biomni was one of few agents to complete all tasks
Training data
Group in Figure 7: Public data + wet lab for validation only.
| Class | Value | Detail |
|---|---|---|
| a. Public databases | Yes | Tools, packages and databases are public resources |
| b. Private or licensed data | Not disclosed | |
| c. Own wet-lab data used for training | No | |
| d. Wet lab used only for validation | Yes | One cloning experiment used for validation source |
Wet-lab validation
Every rate carries its numerator and denominator. Claims that could not be verified are shown as unverified, without numbers; different paper versions are separate entries.
Denominator basis: A single cloning task; success = colonies on selective media plus sequence-verified insertion
System-generated protocols used unmodified; colony counts not given; developer-run, not an independent replication
Background
- Organization: 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
- Products and partnerships: Commercial product Biomni Lab (research preview)
- Related models: Kosmos / Edison Analysis, Robin, 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
- Journal article DOI 10.1126/science.adz4351 (journal)
- bioRxiv preprint (preprint)
- Crossref record 10.1126/science.adz4351 (journal)
- GitHub repository snap-stanford/Biomni (repository)
- Official page pulse2.com (company blog)
- Phylo blog: evaluating AI agents in biology (company blog)
- BioML-bench preprint (third-party paper)
Corrections
None recorded.
Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.