TxBench Preclinical Pharmacology V1 Clean Reruns
Science
Clean-v1 preclinical-pharmacology reruns on 100 experimental-data evaluations. Formerly V2, now V1; results remain separate from the original cohort.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Sonnet 5.5 | 64.3 |
| Claude Opus 5 | 63.0 |
| GPT-6 Sol | 63.0 |
| Grok 4.6 | 61.7 |
| GPT-5.6 Sol | 61.3 |
| GPT-6.1 Sol | 59.7 |
| GPT-6 Astra | 59.3 |
| Opus 4.8 | 57.7 |
| Gemini 3.7 Flash | 55.3 |
| GPT-5.5 | 55.3 |
| Gemini 3.8 Flash | 54.7 |
| Grok 4.7 | 54.7 |
| Sonnet 5 | 49.3 |
| DeepSeek V4.1 Flash | 46.7 |
| Opus 4.7 | 45.3 |
| GPT-6 Luna | 44.3 |
| GPT-5.6 Luna | 43.7 |
| Kimi K3 | 39.7 |
| Sonnet 4.6 | 34.3 |
| Claude Opus 5.5 | 30.7 |
| Nemotron 3.5 Lightning 30B A3B (precision unspecified) | 10.3 |