Calibration
Video-MME (w/o subtitle) → AA-MMMU-Pro
AA-MMMU-Pro is estimated from Video-MME (w/o subtitle) with a Hill curve fitted on 6 models measured on both: y = 0.0000 + (1.0675 − 0.0000)·x^6.00 / (0.72048^6.00 + x^6.00), R² = 0.99, cross-validated error 2.1 pp. It is used for 0 estimates.
Fitted on
| Model | Video-MME (w/o subtitle) | AA-MMMU-Pro |
|---|---|---|
| Claude Opus 4.5 | 81.4% | 71.2% |
| Gemini 3 Pro | 87.7% | 80.2% |
| Kimi K2.5 | 83.2% | 75.4% |
| Nemotron 3 Nano Omni 30B A3B | 72.2% | 53.2% |
| Qwen3.5 397B | 83.7% | 77.3% |
| Qwen3.6-35B-A3B | 82.5% | 75.0% |