Calibration
VideoMMMU → MMMU
MMMU is estimated from VideoMMMU with a Michaelis–Menten curve fitted on 9 models measured on both: y = 2.0000·x / (1.17759 + x), R² = 0.95, cross-validated error 2.4 pp. It is used for 0 estimates.
Fitted on
| Model | VideoMMMU | MMMU |
|---|---|---|
| Claude Opus 4.5 | 84.4% | 80.7% |
| Gemini 3 Pro | 87.6% | 87.2% |
| GPT-5.2 | 85.9% | 86.7% |
| Kimi K2.5 | 86.6% | 84.3% |
| LFM3-VL-2B | 49.6% | 55.7% |
| Qwen3.5 397B | 84.7% | 85.0% |
| Qwen3.6-27B | 84.4% | 82.9% |
| Qwen3.6-35B-A3B | 83.7% | 81.7% |
| Qwen3.6 Plus | 84.0% | 86.0% |