Calibration
SimpleVQA → MMMU
MMMU is estimated from SimpleVQA with a Hill curve fitted on 9 models measured on both: y = 0.0000 + (0.8612 − 0.0000)·x^6.00 / (0.33357^6.00 + x^6.00), R² = 0.97, cross-validated error 3.2 pp. It is used for 0 estimates.
Fitted on
| Model | SimpleVQA | MMMU |
|---|---|---|
| Claude Opus 4.5 | 65.7% | 80.7% |
| Gemini 3 Pro | 73.2% | 87.2% |
| GPT-5.2 | 55.8% | 86.7% |
| Kimi K2.5 | 71.2% | 84.3% |
| LFM2.5-VL-3B | 35.4% | 48.4% |
| LFM3-VL-2B | 36.3% | 55.7% |
| Qwen3.5 397B | 67.1% | 85.0% |
| Qwen3.6-27B | 56.1% | 82.9% |
| Qwen3.6-35B-A3B | 58.9% | 81.7% |