Anthropic
BioMysteryBench
Agentic bioinformatics investigations: recover biological information from raw experimental data using computational analysis.
bioinformaticsmolecular-biologycodetabulartext
Status
curated
Saturation
unsaturated
Reproduction
not yet
Published results
25 results
Editor's note
Agentic bioinformatics investigations using raw biological data. Vals AI reports overall accuracy across 90 tasks, averaged over three full-benchmark runs; these are external evaluations, not BioInspect runs.
Published results
Vals AI · overall · Terminus 2
Mean of three full-benchmark runs. ± is standard error across the three run-level scores, not a 95% confidence interval. Refusals and provider/model failures count as failures. Separate from Anthropic's system-card subset results.
| Model | Score | Variant | Source |
|---|---|---|---|
| Claude Sonnet 5.5 | 81.1% accuracy ± 0.6 pp SE 219/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| GPT-6.1 Sol | 79.6% accuracy ± 0.7 pp SE 215/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| GPT-6 Astra | 79.3% accuracy ± 1.0 pp SE 214/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Claude Opus 5.5 | 79.3% accuracy ± 1.0 pp SE 214/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Claude Opus 5 | 79.3% accuracy ± 2.1 pp SE 214/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Gemini 4 Argon | 76.3% accuracy ± 0.4 pp SE 206/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| GPT-6 Sol | 74.8% accuracy ± 2.4 pp SE 202/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Kimi K3 | 72.6% accuracy ± 1.6 pp SE 196/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Grok 4.6 | 72.2% accuracy ± 0.0 pp SE 195/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| GPT-5.6 Sol | 71.1% accuracy ± 0.6 pp SE 192/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Claude Haiku 5.5 | 70.4% accuracy ± 1.3 pp SE 190/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Grok 4.7 | 69.3% accuracy ± 1.3 pp SE 187/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Hy4 Preview | 69.3% accuracy ± 2.6 pp SE 187/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| MiMo V2.6 Flash | 69.3% accuracy ± 1.0 pp SE 187/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| DeepSeek V4.1 Flash | 67.8% accuracy ± 1.1 pp SE 183/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Step 5 Preview | 67.4% accuracy ± 0.4 pp SE 182/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Ember 1 | 67.0% accuracy ± 3.0 pp SE 181/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Mistral Large 4 | 67.0% accuracy ± 2.7 pp SE 181/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Muse Spark 1.2 | 64.8% accuracy ± 1.3 pp SE 175/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| DeepSeek V4 Flash 0731 | 64.4% accuracy ± 1.3 pp SE 174/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Gemini 3.8 Flash | 62.2% accuracy ± 1.7 pp SE 168/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| GPT-6 Luna | 61.5% accuracy ± 2.6 pp SE 166/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| GPT-5.6 Luna | 61.5% accuracy ± 0.7 pp SE 166/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Gemini 3.1 Pro Preview | 60.0% accuracy ± 2.9 pp SE 162/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |
| Gemini 3.6 Flash | 58.5% accuracy ± 1.6 pp SE 158/270 trials | Overall · 90 tasks × 3 runs · Terminus 2 | Vals AI · BioMysteryBench 2026-10-08 |