benchgap
Calibration

BrowseComp → Toolathlon-Verified

Toolathlon-Verified is estimated from BrowseComp with a linear curve fitted on 10 models measured on both: y = 0.9890·x + -0.1464, R² = 0.71, cross-validated error 7.7 pp. It is used for 7 estimates.

Estimated modelBrowseCompToolathlon-VerifiedSource
Agents-A1-4B66.8%51.4%estimated ± 7.7 pp, medium confidence
GPT-5.4 Pro89.3%73.7%estimated ± 7.7 pp, medium confidence
GPT-5.5 Pro90.1%74.5%estimated ± 7.7 pp, medium confidence
LongCat-Flash-Lite-Sparse48.6%33.4%estimated ± 7.7 pp, low confidence
Qwen3.5-27B61.0%45.7%estimated ± 7.7 pp, medium confidence
Qwen3.5-35B-A3B61.0%45.7%estimated ± 7.7 pp, medium confidence
Beam77.4%61.9%estimated ± 7.7 pp, medium confidence