Calibration

Toolathlon → BrowseComp

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

BrowseComp is estimated from Toolathlon with a Hill curve fitted on 12 models measured on both: y = 0.5780 + (0.9848 − 0.5780)·x^6.00 / (0.48941^6.00 + x^6.00), R² = 0.89, cross-validated error 4.4 pp. It is used for 2 estimates.

Estimated model	Toolathlon	BrowseComp	Source
Qwen3.6-35B-A3B	26.9%	58.9%	estimated ± 4.4 pp, medium confidence
Qwen3.6 Plus	39.8%	66.9%	estimated ± 4.4 pp, high confidence