Calibration
Video-MME (w/o subtitle) → DynaMath
DynaMath is estimated from Video-MME (w/o subtitle) with a offset logistic curve fitted on 6 models measured on both: y = 0.7912 + (0.8616 − 0.7912) / (1 + exp(−200.00·(x − 0.8252))), R² = 0.96, cross-validated error 1.0 pp. It is used for 0 estimates.
Fitted on
| Model | Video-MME (w/o subtitle) | DynaMath |
|---|---|---|
| Claude Opus 4.5 | 81.4% | 79.7% |
| Gemini 3 Pro | 87.7% | 85.1% |
| GPT-5.2 | 85.8% | 86.8% |
| Kimi K2.5 | 83.2% | 84.4% |
| LFM3-VL-2B | 62.8% | 79.2% |
| Qwen3.5 397B | 83.7% | 86.3% |