Calibration
Video-MME (w/o subtitle) → VideoMMMU
VideoMMMU is estimated from Video-MME (w/o subtitle) with a inverse Michaelis–Menten curve fitted on 7 models measured on both: y = 1.53426·(x − 0.1269) / (0.1269 + 2.0000 − x), R² = 0.95, cross-validated error 8.9 pp. It is used for 0 estimates.
Fitted on
| Model | Video-MME (w/o subtitle) | VideoMMMU |
|---|---|---|
| Claude Opus 4.5 | 81.4% | 84.4% |
| Gemini 3 Pro | 87.7% | 87.6% |
| GPT-5.2 | 85.8% | 85.9% |
| Kimi K2.5 | 83.2% | 86.6% |
| LFM3-VL-2B | 62.8% | 49.6% |
| Qwen3.5 397B | 83.7% | 84.7% |
| Qwen3.6-35B-A3B | 82.5% | 83.7% |