benchgap
Calibration

Artificial Analysis Intelligence Index → MMLU-Pro

MMLU-Pro is estimated from Artificial Analysis Intelligence Index with a Hill curve fitted on 36 models measured on both: y = 0.7154 + (0.8691 − 0.7154)·x^6.00 / (0.13703^6.00 + x^6.00), R² = 0.67, cross-validated error 4.5 pp. It is used for 2 estimates.

Estimated modelArtificial Analysis Intelligence IndexMMLU-ProSource
Claude 4.1 Opus18.6%84.8%estimated ± 4.5 pp, high confidence
GPT-4 Turbo7.0%71.8%estimated ± 4.5 pp, high confidence