benchgap
Calibration

Video-MME (with subtitle) → Video-MME (w/o subtitle)

Video-MME (w/o subtitle) is estimated from Video-MME (with subtitle) with a Michaelis–Menten curve fitted on 7 models measured on both: y = 2.0000·x / (1.21733 + x), R² = 0.70, cross-validated error 6.2 pp. It is used for 0 estimates.

Fitted on

ModelVideo-MME (with subtitle)Video-MME (w/o subtitle)
Claude Opus 4.577.6%81.4%
Gemini 3 Pro88.4%87.7%
GPT-5.286.0%85.8%
Kimi K2.587.4%83.2%
LFM3-VL-2B69.0%62.8%
Qwen3.5 397B87.5%83.7%
Qwen3.6-35B-A3B86.6%82.5%