benchgap
Calibration

BrowseComp → GDPval-AA

GDPval-AA is estimated from BrowseComp with a Hill curve fitted on 34 models measured on both: y = 0.1798 + (1.2000 − 0.1798)·x^6.00 / (1.00344^6.00 + x^6.00), R² = 0.73, cross-validated error 7.8 pp. It is used for 10 estimates.

Estimated modelBrowseCompGDPval-AASource
GPT-5.5 Pro90.1%53.1%estimated ± 7.8 pp, medium confidence
GPT-5.4 Pro89.3%51.8%estimated ± 7.8 pp, medium confidence
Nemotron 3.5 Bolt 74B-A9B BF1670.0%28.5%estimated ± 7.8 pp, medium confidence
Qwen3.5 397B69.0%27.7%estimated ± 7.8 pp, medium confidence
Agents-A1-4B66.8%26.2%estimated ± 7.8 pp, medium confidence
Qwen3.5-27B61.0%22.9%estimated ± 7.8 pp, medium confidence
Qwen3.5-35B-A3B61.0%22.9%estimated ± 7.8 pp, medium confidence
Sarvam 105B49.5%19.4%estimated ± 7.8 pp, medium confidence
LongCat-Flash-Lite-Sparse48.6%19.3%estimated ± 7.8 pp, medium confidence
Sarvam 30B35.5%18.2%estimated ± 7.8 pp, low confidence

Fitted on

ModelBrowseCompGDPval-AA
Claude Opus 4.779.3%42.6%
Claude Opus 4.884.3%47.7%
Claude Opus 590.8%61.1%
Claude Sonnet 584.7%48.2%
DeepSeek V4 Flash 073173.2%46.7%
DeepSeek V4 Pro83.4%54.5%
GLM-4.752.0%25.7%
GLM-5.168.0%30.8%
GPT-5.482.7%37.3%
GPT-5.584.4%42.5%
GPT-5.6 Luna83.3%48.0%
GPT-5.6 Sol92.2%55.4%
GPT-5.6 Terra87.5%47.6%
GPT-6 Astra91.5%53.7%
Inkling77.1%28.6%
Inkling-Small77.4%31.2%
Kimi K2.683.2%26.9%
Kimi K3.1 Pro92.1%67.3%
Kimi K2.560.6%17.0%
Kimi K391.2%51.6%
Ling 3.0 Flash72.2%22.4%
Ling 3.1 Flash91.7%55.9%
MiniMax M383.5%37.1%
Mistral Large 484.5%46.1%
Nemotron 3.5 Bolt 74B-A9B66.8%24.0%
Nemotron 3.5 Lightning 30B A3B NVFP436.8%6.9%
Nemotron 3.5 Super VL 120B A12B78.0%33.1%
Nemotron 3.5 Super VL 120B A12B NVFP477.5%33.1%
Nemotron 3 Ultra44.4%33.1%
Qwen3.5-122B-A10B63.8%15.7%
Beam 1.0 501B-A23B76.8%33.9%
Solar Pro 449.2%30.3%
Step 3.7 Flash75.8%25.8%
Step 5 Preview88.7%54.3%