benchgap
Calibration

MathVision → MMMU-Pro

MMMU-Pro is estimated from MathVision with a Michaelis–Menten curve fitted on 13 models measured on both: y = 2.0000·x / (1.35101 + x), R² = 0.75, cross-validated error 2.2 pp. It is used for 3 estimates.

Estimated modelMathVisionMMMU-ProSource
Qwen3.5-122B-A10B86.2%77.9%estimated ± 2.2 pp, high confidence
Qwen3.5-27B86.0%77.8%estimated ± 2.2 pp, high confidence
Qwen3.5-35B-A3B83.9%76.6%estimated ± 2.2 pp, high confidence