SpatialBench-Long V2
Science
SpatialBench-Long V2 clean-v1 reruns on 21 long-horizon spatial-transcriptomics evaluations. Macro-mean pass rate; the aggregate site's weight of 24 is not the task count.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 41.3 |
| GPT-6 Astra | 39.7 |
| GPT-5.6 Sol | 36.5 |
| GPT-5.5 | 34.9 |
| Opus 4.8 | 33.3 |
| Opus 4.7 | 30.2 |
| GPT-6 Sol | 28.6 |
| Grok 4.6 | 27.0 |
| Gemini 3.7 Flash | 25.4 |
| Sonnet 4.6 | 23.8 |
| Sonnet 5 | 23.8 |
| Gemini 3.8 Flash | 23.8 |
| GPT-5.6 Luna | 23.8 |
| GPT-6 Luna | 23.8 |
| Grok 4.7 | 23.8 |
| Claude Opus 5.5 | 19.1 |