Calibration

MMMU → AA-MMMU-Pro

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

AA-MMMU-Pro is estimated from MMMU with a offset logistic curve fitted on 8 models measured on both: y = 0.1307 + (0.8226 − 0.1307) / (1 + exp(−15.12·(x − 0.6867))), R² = 0.99, cross-validated error 1.2 pp. It is used for 3 estimates.

Estimated model	MMMU	AA-MMMU-Pro	Source
LFM2.5-VL-3B	48.4%	16.2%	estimated ± 1.2 pp, medium confidence
LFM2.5-VL-450M	32.7%	13.4%	estimated ± 1.2 pp, medium confidence
ZAYA1-VL-8B	46.0%	15.2%	estimated ± 1.2 pp, medium confidence