Claw-Eval General (English) average over three trials
Agents · 2026-04-07
Mean of Claw-Eval's continuous per-task scores, which combine completion and robustness under a safety gate, over three trials of the English General-split tasks. It comes from the same runs as the pass^3 metric and shares its factor group, so the two do not count as independent evidence.
Top models (higher is better)
| Model | Score |
|---|---|
| LFM3-1.7B | 68.6 |
| Qwen3.5-9B | 66.5 |
| LFM2.5-2.6B | 62.6 |
| Qwen3.5-4B | 62.0 |