ScBench-Long V2
Science
ScBench-Long V2 clean-v1 reruns on 22 long-horizon single-cell analysis evaluations. Macro-mean pass rate; the aggregate site's weight of 21 is not the task count.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 34.9 |
| Claude Opus 5 | 30.3 |
| GPT-6 Astra | 25.8 |
| Opus 4.8 | 24.2 |
| Gemini 3.8 Flash | 24.2 |
| GPT-5.5 | 24.2 |
| GPT-6 Sol | 24.2 |
| Gemini 3.7 Flash | 19.7 |
| Grok 4.6 | 19.7 |
| Sonnet 5 | 15.2 |
| GPT-5.6 Luna | 15.2 |
| Opus 4.7 | 12.1 |
| Sonnet 4.6 | 12.1 |
| GPT-6 Luna | 10.6 |
| Grok 4.7 | 9.1 |
| Claude Opus 5.5 | 7.6 |