Benchmarks / Agentic
BioMysteryBench
Anthropic's benchmark of whether AI agents can solve real-world bioinformatics mysteries from raw data
BioMysteryBench
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Sonnet 5.5 | max | 81.1% |
| 2 | GPT-6.1 Sol | max | 79.6% |
| 3 | GPT-6 Astra | max | 79.3% |
| 4 | Opus 5.5 | max | 79.3% |
| 5 | Opus 5 | max | 79.3% |
| 6 | Gemini 4 | high | 76.3% |
| 7 | GPT-6 Sol | max | 74.8% |
| 8 | Wonder Full | max | 73.3% |
| 9 | Kimi K3 | max | 72.6% |
| 10 | Grok 4.6 | high | 72.2% |
| 11 | GPT-5.6 Sol | max | 71.1% |
| 12 | Haiku 5.5 | max | 70.4% |
| 13 | Grok 4.7 | xhigh | 69.3% |
| 14 | Hy4 | – | 69.3% |
| 15 | MiMo V2.6 Flash | high | 69.3% |
| 16 | DeepSeek V4.1 Flash | high | 67.8% |
| 17 | Step 5 Preview | – | 67.4% |
| 18 | Ember-1 | – | 67.0% |
| 19 | Wonder | max | 67.0% |
| 20 | Mistral Large 4 | high | 67.0% |
| 21 | Muse Spark 1.2 | xhigh | 64.8% |
| 22 | DeepSeek V4 Flash 0731 | high | 64.4% |
| 23 | Gemini 3.8 Flash | high | 62.2% |
| 24 | GPT-6 Luna | max | 61.5% |
| 25 | GPT-5.6 Luna | max | 61.5% |
| 26 | Gemini 3.1 Pro | high | 60.0% |
| 27 | Gemini 3.6 Flash | high | 58.5% |