MetagenomicsBench V1
Science
MetagenomicsBench component of benchmarks.bio's current V1 clean-rerun export, on 100 evaluations. Published macro-mean pass rates distinguish model and agent harness; missing scores remain missing.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 60.0 |
| Claude Opus 5 | 54.0 |
| Opus 4.8 | 50.0 |
| GPT-6 Astra | 49.7 |
| Grok 4.7 | 48.7 |
| Claude Opus 5.5 | 43.7 |
| Sonnet 5 | 42.3 |
| GPT-6 Sol | 42.0 |
| Grok 4.6 | 42.0 |
| DeepSeek V4.1 Flash | 38.0 |
| Opus 4.7 | 35.3 |
| Gemini 3.7 Flash | 31.7 |
| GPT-5.6 Sol | 31.3 |
| GPT-6 Luna | 30.7 |
| Sonnet 4.6 | 28.3 |
| Kimi K3 | 27.3 |
| Gemini 3.8 Flash | 26.7 |
| GPT-5.6 Luna | 24.3 |
| Nemotron 3.5 Lightning 30B A3B (precision unspecified) | 7.7 |