Calibration
VideoMMMU → Video-MME (with subtitle)
Video-MME (with subtitle) is estimated from VideoMMMU with a Michaelis–Menten + offset curve fitted on 11 models measured on both: y = 0.3569 + 2.0000·x / (2.51135 + x), R² = 0.77, cross-validated error 3.9 pp. It is used for 0 estimates.
Fitted on
| Model | VideoMMMU | Video-MME (with subtitle) |
|---|---|---|
| Claude Opus 4.5 | 84.4% | 77.6% |
| Gemini 3 Pro | 87.6% | 88.4% |
| GPT-5.2 | 85.9% | 86.0% |
| Kimi K2.5 | 86.6% | 87.4% |
| LFM3-VL-2B | 49.6% | 69.0% |
| MiniMax M3 | 84.6% | 85.4% |
| Qwen3.5 397B | 84.7% | 87.5% |
| Qwen3.6-27B | 84.4% | 87.7% |
| Qwen3.6-35B-A3B | 83.7% | 86.6% |
| Qwen3.7 Plus | 85.4% | 88.0% |
| Qwen3.8 Max | 88.7% | 90.4% |