Scientific Research Automation on MetaClaw-Bench Part II (14 days, 588 Q)
49.1AccuracyMetaClaw (Skills)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MetaClaw (Skills)Model=GPT-5.22026.03 | 49.1 | 67.5 | |
| GPT-5.2Condition=Baseline2026.03 | 44.9 | 58.4 | |
| MetaClaw (Full)Model=Kimi-K2.52026.03 | 39.6 | 51.9 | |
| MetaClaw (Skills)Model=Kimi-K2.52026.03 | 26.9 | 33.8 | |
| Kimi-K2.5Condition=Baseline2026.03 | 21.1 | 18.2 |