Calibration
MathVision → CharXiv
CharXiv is estimated from MathVision with a Hill curve fitted on 15 models measured on both: y = 0.5183 + (1.2000 − 0.5183)·x^6.00 / (0.90108^6.00 + x^6.00), R² = 0.78, cross-validated error 3.3 pp. It is used for 3 estimates.
| Estimated model | MathVision | CharXiv | Source |
|---|---|---|---|
| Gemma 4 12B | 79.7% | 73.9% | estimated ± 3.3 pp, high confidence |
| Qwen3.5-27B | 86.0% | 81.2% | estimated ± 3.3 pp, high confidence |
| Qwen3.5-35B-A3B | 83.9% | 78.7% | estimated ± 3.3 pp, high confidence |