Overview
| Name | Robin |
|---|---|
| Full name | No separate expanded name |
| Ambiguous names (not auto-linked) | Robin |
| Family | Not a family entry. Single-model entry, not a family |
| Organization | FutureHouse |
| Organization type | Nonprofit institute |
| Country | United States |
| Model class | Multi-agent discovery system |
| Domain | Text, LLMs & agents |
| First release | · preprint Basis: arXiv 2505.13400 v1 submitted source |
| Other milestones | |
| Latest update | · journal Basis: Nature journal publication Paper published in Nature source The most recent sourced event, not a claim that nothing changed since. |
| Status | Active · Journal publication within 12 months |
| Openness | Code: Open Weights: Unclear Code released, but running it needs a FutureHouse platform key |
| Benchmarks reported | 0 counted · none of the 44 |
| Wet-lab validation | 1 recorded campaign |
| Unverified fields | None |
| As of | 2026-11-16 |
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.
Benchmarks reported
None of the 44 benchmarks. This model is on the zero-pairing list.
How it evaluates itself: Dry AMD lab-in-the-loop; Effect size conflict
Other evaluations (outside the 44)
- Dry-AMD lab-in-the-loop (Wet-lab validation) Self-reported: Two rounds of cell assays: 5 of 30 candidates, then 10
- Effect size conflict (Case study) Self-reported: Same effect: agent analysis 7.5x, human 1.75x; authors call for more testing
Developers report: Developers report the first multi-agent system to autonomously hypothesize, plan, analyze and propose a validated new candidate
Independent evidence: No independent evidence found. No independent evidence found
Training data
Group in Figure 7: Public data + wet lab for validation only.
| Class | Value | Detail |
|---|---|---|
| a. Public databases | Yes | Base model and retrieval tools use public literature |
| b. Private or licensed data | Not disclosed | |
| c. Own wet-lab data used for training | No | |
| d. Wet lab used only for validation | Yes | Cell phagocytosis assays tested candidates 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: Round 1 tested 5 of 30 proposed candidates, round 2 tested 10; no hit rate reported
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
Background
- Organization: FutureHouse is a nonprofit AI-science lab founded in San Francisco in 2023-09; its commercial arm is Edison Scientific
- Funding: Not disclosed or not applicable
- Products and partnerships: Done with Oxford collaborators; related products Kosmos and Finch
- Related models: Kosmos / Edison Analysis, 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
- Crossref record 10.1038/s41586-026-10652-y (journal)
- Official page futurehouse.org (company blog)
- arXiv preprint 2505.13400 (preprint)
- Journal or proceedings page (journal)
Corrections
None recorded.
Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.