BioLP-bench
Science · 2024-08-21
BioLP-bench asks models to find the deliberately introduced mistake in modified biological laboratory protocols, measuring practical protocol-troubleshooting knowledge. Scores report accuracy; the authors report human experts at about 38%.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4 | 47.0 |
| Grok 4 Fast | 39.0 |
| Grok 4.1 | 37.0 |
| Grok Code Fast 1 | 19.9 |