BioSecBench-Function V1 Clean Reruns
Science
Clean-v1 reruns on 111 deterministic biological-function evaluations. Formerly V2, now V1; macro-mean pass rates on updated evaluation sets remain separate from the original cohort.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 73.6 |
| Claude Sonnet 5.5 | 53.5 |
| GPT-6 Astra | 48.6 |
| GPT-6.1 Sol | 46.0 |
| Claude Opus 5 | 44.8 |
| Grok 4.6 | 44.0 |
| Grok 4.7 | 43.3 |
| Opus 4.8 | 43.1 |
| DeepSeek V4.1 Flash | 40.8 |
| GPT-6 Sol | 39.3 |
| Gemini 3.8 Flash | 36.0 |
| GPT-5.6 Sol | 35.5 |
| Sonnet 5 | 33.8 |
| Gemini 3.7 Flash | 33.7 |
| Kimi K3 | 32.7 |
| Opus 4.7 | 30.8 |
| GPT-5.5 | 28.3 |
| GPT-5.6 Luna | 27.6 |
| Sonnet 4.6 | 27.3 |
| GPT-6 Luna | 27.3 |
| Nemotron 3.5 Lightning 30B A3B (precision unspecified) | 13.8 |