Calibration

Video-MME (with subtitle) → MLVU (M-Avg)

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

MLVU (M-Avg) is estimated from Video-MME (with subtitle) with a Michaelis–Menten + offset curve fitted on 9 models measured on both: y = 0.2170 + 2.0000·x / (1.83197 + x), R² = 0.52, cross-validated error 2.1 pp. It is used for 1 estimate.

Estimated model	Video-MME (with subtitle)	MLVU (M-Avg)	Source
MiniMax M3	85.4%	85.3%	estimated ± 2.1 pp, high confidence