Calibration
Video-MME (with subtitle) → CountBench
CountBench is estimated from Video-MME (with subtitle) with a offset logistic curve fitted on 6 models measured on both: y = 0.9075 + (0.9820 − 0.9075) / (1 + exp(−200.00·(x − 0.8701))), R² = 0.88, cross-validated error 1.8 pp. It is used for 2 estimates.
| Estimated model | Video-MME (with subtitle) | CountBench | Source |
|---|---|---|---|
| MiMo-V2.5 | 87.7% | 96.7% | estimated ± 1.8 pp, medium confidence |
| MiniMax M3 | 85.4% | 91.0% | estimated ± 1.8 pp, medium confidence |