benchgap
Calibration

MMMU → MMMU-Pro

MMMU-Pro is estimated from MMMU with a inverse Michaelis–Menten curve fitted on 5 models measured on both: y = 1.36258·(x − 0.1166) / (0.1166 + 2.0000 − x), R² = 1.00, cross-validated error 2.4 pp. It is used for 3 estimates.

Estimated modelMMMUMMMU-ProSource
LFM2.5-VL-450M32.7%16.0%estimated ± 2.4 pp, low confidence
Nemotron 3 Nano Omni 30B A3B70.8%57.2%estimated ± 2.4 pp, medium confidence
ZAYA1-VL-8B46.0%28.2%estimated ± 2.4 pp, low confidence