benchgap
Calibration

HLE w/ tools → BrowseComp

BrowseComp is estimated from HLE w/ tools with a Michaelis–Menten + offset curve fitted on 12 models measured on both: y = 0.0000 + 2.0000·x / (0.78375 + x), R² = 0.75, cross-validated error 7.6 pp. It is used for 1 estimate.

Estimated modelHLE w/ toolsBrowseCompSource
Claude Haiku 5.557.4%84.6%estimated ± 7.6 pp, medium confidence