Calibration
Claw-Eval → BrowseComp
BrowseComp is estimated from Claw-Eval with a Michaelis–Menten curve fitted on 12 models measured on both: y = 2.0000·x / (1.11765 + x), R² = 0.33, cross-validated error 9.4 pp. It is used for 8 estimates.
| Estimated model | Claw-Eval | BrowseComp | Source |
|---|---|---|---|
| GLM-5-Turbo | 55.8% | 66.6% | estimated ± 9.4 pp, low confidence |
| K-EXAONE 2.0 | 77.7% | 82.0% | estimated ± 9.4 pp, low confidence |
| LFM2.5-2.6B | 62.8% | 72.0% | estimated ± 9.4 pp, low confidence |
| LLaDA2.2-mini | 57.2% | 67.7% | estimated ± 9.4 pp, low confidence |
| MiMo-V2-Omni | 45.2% | 57.6% | estimated ± 9.4 pp, low confidence |
| Ornith-1.0-35B | 69.8% | 76.9% | estimated ± 9.4 pp, low confidence |
| Ornith-1.0-397B | 77.1% | 81.6% | estimated ± 9.4 pp, low confidence |
| Ornith-1.0-9B | 63.1% | 72.2% | estimated ± 9.4 pp, low confidence |