benchgap
Calibration

Video-MME (with subtitle) → MMMU-Pro

MMMU-Pro is estimated from Video-MME (with subtitle) with a inverse Michaelis–Menten curve fitted on 12 models measured on both: y = 1.48425·(x − 0.1810) / (0.1810 + 2.0000 − x), R² = 0.88, cross-validated error 4.2 pp. It is used for 0 estimates.

Fitted on

ModelVideo-MME (with subtitle)MMMU-Pro
Claude Opus 4.577.6%70.6%
Gemini 3 Pro88.4%81.0%
GPT-5.286.0%79.5%
Kimi K2.587.4%78.5%
LFM3-VL-2B69.0%45.7%
MiMo-V2.587.7%77.9%
MiniMax M385.4%78.1%
Qwen3.5 397B87.5%79.0%
Qwen3.6-27B87.7%75.8%
Qwen3.6-35B-A3B86.6%75.3%
Qwen3.7 Plus88.0%79.0%
Qwen3.8 Max90.4%82.3%