Calibration
MathVision → CharXiv w/o tools
CharXiv w/o tools is estimated from MathVision with a offset logistic curve fitted on 6 models measured on both: y = 0.8372 + (0.9761 − 0.8372) / (1 + exp(−200.00·(x − 0.9554))), R² = 0.94, cross-validated error 1.7 pp. It is used for 13 estimates.
| Estimated model | MathVision | CharXiv w/o tools | Source |
|---|---|---|---|
| Claude Opus 4.5 | 74.3% | 83.7% | estimated ± 1.7 pp, low confidence |
| Gemini 3 Pro | 86.6% | 83.7% | estimated ± 1.7 pp, low confidence |
| Gemma 4 12B | 79.7% | 83.7% | estimated ± 1.7 pp, low confidence |
| GPT-5.2 | 83.0% | 83.7% | estimated ± 1.7 pp, low confidence |
| Kimi K2.6 | 87.4% | 83.7% | estimated ± 1.7 pp, low confidence |
| Qwen3.5-122B-A10B | 86.2% | 83.7% | estimated ± 1.7 pp, low confidence |
| Qwen3.5-27B | 86.0% | 83.7% | estimated ± 1.7 pp, low confidence |
| Qwen3.5-35B-A3B | 83.9% | 83.7% | estimated ± 1.7 pp, low confidence |
| Qwen3.5 397B | 88.6% | 83.7% | estimated ± 1.7 pp, medium confidence |
| Qwen3.6 Plus | 88.0% | 83.7% | estimated ± 1.7 pp, medium confidence |
| Qwen3.7 Plus | 90.3% | 83.7% | estimated ± 1.7 pp, medium confidence |
| Seed 2.1 Pro | 92.6% | 83.8% | estimated ± 1.7 pp, medium confidence |
| Seed 2.1 Turbo | 90.1% | 83.7% | estimated ± 1.7 pp, medium confidence |