Calibration
VideoMMMU → Video-MME (w/o subtitle)
Video-MME (w/o subtitle) is estimated from VideoMMMU with a inverse Michaelis–Menten curve fitted on 6 models measured on both: y = 1.12536·(x − 0.0000) / (0.0000 + 2.0000 − x), R² = 0.53, cross-validated error 1.7 pp. It is used for 4 estimates.
| Estimated model | VideoMMMU | Video-MME (w/o subtitle) | Source |
|---|---|---|---|
| Qwen3.8 Max | 88.7% | 89.7% | estimated ± 1.7 pp, low confidence |
| dots3-note Preview | 86.8% | 86.3% | estimated ± 1.7 pp, medium confidence |
| Qwen3.7 Plus | 85.4% | 83.9% | estimated ± 1.7 pp, medium confidence |
| MiniMax M3 | 84.6% | 82.5% | estimated ± 1.7 pp, medium confidence |