Vals AI BioMysteryBench: leaderboard
BioMysteryBench (Vals AI) — end-to-end bioinformatics investigations where the agent must recover hidden metadata from raw biological data through its own computational analysis, spanning sequencing, expression, variant calling, epigenomics, metagenomics, proteomics and metabolomics; accuracy over the private task set.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around March 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5.5 | 81.11 | #11 |
| 2 | GPT-6.1 Sol | 79.63 | #7 |
| 3 | GPT-6 Astra | 79.26 | #6 |
| 4 | Claude Opus 5 (Max) | 79.26 | #12 (Claude Opus 5) |
| 5 | Claude Opus 5.5 | 79.26 | #3 |
| 6 | Gemini 4 Argon | 76.3 | #8 |
| 7 | GPT-6 Sol | 74.81 | #14 |
| 8 | Muse Wonder Full | 73.33 | #10 |
| 9 | Kimi K3 | 72.59 | #30 |
| 10 | Grok 4.6 | 72.22 | #26 |
| 11 | GPT-5.6 Sol (Max) | 71.11 | #19 (GPT-5.6 Sol) |
| 12 | Claude Haiku 5.5 | 70.37 | #40 |
| 13 | MiMo-V2.6-Flash | 69.26 | #70 |
| 14 | Grok 4.7 | 69.26 | #25 |
| 15 | Hy4 preview | 69.26 | #38 |
No result here: #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-biomysterybench · How It Works · Data refreshed daily, snapshot 2026-10-20.