Calibration
VideoMMMU → RefCOCO (avg)
RefCOCO (avg) is estimated from VideoMMMU with a linear curve fitted on 5 models measured on both: y = -2.1698·x + 2.7504, R² = 0.91, cross-validated error 1.9 pp. It is used for 3 estimates.
| Estimated model | VideoMMMU | RefCOCO (avg) | Source |
|---|---|---|---|
| Qwen3.6 Plus | 84.0% | 92.8% | estimated ± 1.9 pp, medium confidence |
| MiniMax M3 | 84.6% | 91.5% | estimated ± 1.9 pp, medium confidence |
| dots3-note Preview | 86.8% | 86.7% | estimated ± 1.9 pp, medium confidence |