Calibration
SimpleVQA → MathVision
MathVision is estimated from SimpleVQA with a Michaelis–Menten curve fitted on 9 models measured on both: y = 2.0000·x / (0.93224 + x), R² = 0.82, cross-validated error 7.2 pp. It is used for 0 estimates.
Fitted on
| Model | SimpleVQA | MathVision |
|---|---|---|
| Claude Opus 4.5 | 65.7% | 74.3% |
| dots3-note Preview | 72.5% | 87.7% |
| Gemini 3 Pro | 73.2% | 86.6% |
| GPT-5.2 | 55.8% | 83.0% |
| Kimi K2.5 | 71.2% | 84.2% |
| LFM3-VL-2B | 36.3% | 51.6% |
| Qwen3.5 397B | 67.1% | 88.6% |
| Qwen3.7 Plus | 81.7% | 90.3% |
| Qwen3.8 Max | 75.0% | 95.2% |