Scientific Research Automation on MetaClaw-Bench Part I (30 days, 346 Q)
44AccuracyMetaClaw (Skills)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MetaClaw (Skills)Model=GPT-5.22026.03 | 44 | 17.1 | |
| GPT-5.2Condition=Baseline2026.03 | 41.1 | 14.7 | |
| MetaClaw (Full)Model=Kimi-K2.52026.03 | 40.6 | 16.5 | |
| MetaClaw (Skills)Model=Kimi-K2.52026.03 | 28.3 | 2 | |
| Kimi-K2.5Condition=Baseline2026.03 | 21.4 | 2 |