EpiBench V1 Clean Reruns
Science
Clean-v1 reruns on 106 epigenomics 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 | 39.0 |
| Claude Sonnet 5.5 | 30.2 |
| GPT-5.5 | 28.3 |
| GPT-6.1 Sol | 28.3 |
| GPT-6 Astra | 28.3 |
| Opus 4.8 | 25.8 |
| GPT-5.6 Sol | 25.8 |
| DeepSeek V4.1 Flash | 25.5 |
| Claude Opus 5 | 25.2 |
| Grok 4.6 | 23.6 |
| Opus 4.7 | 22.6 |
| Gemini 3.7 Flash | 22.0 |
| GPT-6 Sol | 22.0 |
| Sonnet 5 | 21.7 |
| Gemini 3.8 Flash | 21.4 |
| Grok 4.7 | 20.8 |
| GPT-5.6 Luna | 19.8 |
| GPT-6 Luna | 19.2 |
| Sonnet 4.6 | 18.9 |
| Kimi K3 | 15.7 |
| Nemotron 3.5 Lightning 30B A3B (precision unspecified) | 6.9 |