benchgap
Calibration

SimpleQA → MMLU-Pro

MMLU-Pro is estimated from SimpleQA with a Michaelis–Menten + offset curve fitted on 7 models measured on both: y = 0.0000 + 0.8948·x / (0.03317 + x), R² = 0.75, cross-validated error 4.8 pp. It is used for 1 estimate.

Estimated modelSimpleQAMMLU-ProSource
Gemini 2.0 Flash-Lite16.5%74.5%estimated ± 4.8 pp, medium confidence

Fitted on

ModelSimpleQAMMLU-Pro
DeepSeek-R130.1%84.0%
DeepSeek-R1-052827.8%85.0%
DeepSeek V324.9%75.9%
Grok 3 [Beta]43.6%79.9%
Kimi K231.0%81.1%
Mistral Small 3.110.4%66.8%
Qwen3 235B 250754.3%83.0%