Calibration
Video-MME (w/o subtitle) → AI2D_TEST
AI2D_TEST is estimated from Video-MME (w/o subtitle) with a Michaelis–Menten + offset curve fitted on 7 models measured on both: y = 0.5598 + 2.0000·x / (3.82223 + x), R² = 0.50, cross-validated error 2.5 pp. It is used for 0 estimates.
Fitted on
| Model | Video-MME (w/o subtitle) | AI2D_TEST |
|---|---|---|
| Claude Opus 4.5 | 81.4% | 87.7% |
| Gemini 3 Pro | 87.7% | 94.1% |
| GPT-5.2 | 85.8% | 92.2% |
| Kimi K2.5 | 83.2% | 90.8% |
| Nemotron 3 Nano Omni 30B A3B | 72.2% | 88.5% |
| Qwen3.5 397B | 83.7% | 93.9% |
| Qwen3.6-35B-A3B | 82.5% | 92.7% |