Calibration
RealWorldQA → MMMU
MMMU is estimated from RealWorldQA with a inverse Michaelis–Menten curve fitted on 5 models measured on both: y = 0.36304·(x − 0.0000) / (0.0000 + 1.2240 − x), R² = 0.97, cross-validated error 6.6 pp. It is used for 3 estimates.
| Estimated model | RealWorldQA | MMMU | Source |
|---|---|---|---|
| North Micro Vision Instruct | 62.2% | 37.5% | estimated ± 6.6 pp, low confidence |
| Qwen3.8-Omni-Flash | 87.7% | 91.7% | estimated ± 6.6 pp, low confidence |
| Ternary Bonsai 2 27B | 80.1% | 68.8% | estimated ± 6.6 pp, low confidence |