Calibration
ApprenticeBench → BrowseComp
BrowseComp is estimated from ApprenticeBench with a Hill curve fitted on 10 models measured on both: y = 0.8313 + (0.9151 − 0.8313)·x^5.21 / (0.17712^5.21 + x^5.21), R² = 0.70, cross-validated error 2.8 pp. It is used for 2 estimates.
| Estimated model | ApprenticeBench | BrowseComp | Source |
|---|---|---|---|
| Claude Fable 5 | 34.0% | 91.2% | estimated ± 2.8 pp, high confidence |
| Grok 4.6 | 13.0% | 84.5% | estimated ± 2.8 pp, high confidence |