benchgap
Calibration

MMMU → V*

V* is estimated from MMMU with a Michaelis–Menten curve fitted on 5 models measured on both: y = 2.0000·x / (0.93466 + x), R² = 0.66, cross-validated error 1.4 pp. It is used for 6 estimates.

Estimated modelMMMUV*Source
Command A+75.1%89.1%estimated ± 1.4 pp, low confidence
LFM2.5-VL-3B48.4%68.2%estimated ± 1.4 pp, low confidence
LFM2.5-VL-450M32.7%51.8%estimated ± 1.4 pp, low confidence
Nemotron 3 Nano Omni 30B A3B70.8%86.2%estimated ± 1.4 pp, low confidence
Qwen3.6-35B-A3B81.7%93.3%estimated ± 1.4 pp, medium confidence
ZAYA1-VL-8B46.0%66.0%estimated ± 1.4 pp, low confidence