Calibration
Video-MME (with subtitle) → ERQA
ERQA is estimated from Video-MME (with subtitle) with a offset logistic curve fitted on 8 models measured on both: y = 0.4739 + (0.8660 − 0.4739) / (1 + exp(−47.23·(x − 0.8776))), R² = 0.97, cross-validated error 2.9 pp. It is used for 3 estimates.
| Estimated model | Video-MME (with subtitle) | ERQA | Source |
|---|---|---|---|
| MiMo-V2.5 | 87.7% | 66.7% | estimated ± 2.9 pp, high confidence |
| Qwen3.6-35B-A3B | 86.6% | 61.8% | estimated ± 2.9 pp, high confidence |
| MiniMax M3 | 85.4% | 57.1% | estimated ± 2.9 pp, high confidence |
Fitted on
| Model | Video-MME (with subtitle) | ERQA |
|---|---|---|
| Claude Opus 4.5 | 77.6% | 46.8% |
| Gemini 3 Pro | 88.4% | 70.5% |
| GPT-5.2 | 86.0% | 59.8% |
| LFM3-VL-2B | 69.0% | 48.2% |
| Qwen3.5 397B | 87.5% | 67.5% |
| Qwen3.6-27B | 87.7% | 62.5% |
| Qwen3.7 Plus | 88.0% | 69.8% |
| Qwen3.8 Max | 90.4% | 77.8% |