Calibration
CharXiv w/o tools → MathVision
MathVision is estimated from CharXiv w/o tools with a Hill curve fitted on 6 models measured on both: y = 0.0000 + (1.1299 − 0.0000)·x^6.00 / (0.66390^6.00 + x^6.00), R² = 0.64, cross-validated error 2.1 pp. It is used for 8 estimates.
| Estimated model | CharXiv w/o tools | MathVision | Source |
|---|---|---|---|
| Claude Mythos 5 | 88.9% | 96.3% | estimated ± 2.1 pp, low confidence |
| Claude Opus 4.7 (Adaptive) | 82.1% | 88.3% | estimated ± 2.1 pp, low confidence |
| Claude Opus 4.8 | 80.5% | 86.0% | estimated ± 2.1 pp, low confidence |
| Claude Sonnet 5 | 77.0% | 80.1% | estimated ± 2.1 pp, low confidence |
| Gemini 3.7 Flash | 84.5% | 91.5% | estimated ± 2.1 pp, medium confidence |
| Gemini 3.8 Flash | 86.2% | 93.5% | estimated ± 2.1 pp, medium confidence |
| Inkling | 78.1% | 82.0% | estimated ± 2.1 pp, low confidence |
| Inkling-Small | 77.4% | 80.8% | estimated ± 2.1 pp, low confidence |