Calibration
CharXiv → MathVision w/ Python
MathVision w/ Python is estimated from CharXiv with a offset logistic curve fitted on 5 models measured on both: y = 0.7681 + (0.9770 − 0.7681) / (1 + exp(−138.50·(x − 0.8894))), R² = 0.80, cross-validated error 0.9 pp. It is used for 24 estimates.
| Estimated model | CharXiv | MathVision w/ Python | Source |
|---|---|---|---|
| Claude Mythos 5 | 93.5% | 97.7% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.7 (Adaptive) | 91.0% | 96.6% | estimated ± 0.9 pp, medium confidence |
| Claude Opus 4.8 | 89.9% | 93.3% | estimated ± 0.9 pp, low confidence |
| Claude Sonnet 4.6 | 77.4% | 76.8% | estimated ± 0.9 pp, low confidence |
| Claude Sonnet 5 | 88.3% | 82.9% | estimated ± 0.9 pp, low confidence |
| Command A+ | 52.7% | 76.8% | estimated ± 0.9 pp, low confidence |
| Gemini 3.1 Flash-Lite | 73.2% | 76.8% | estimated ± 0.9 pp, low confidence |
| Gemini 3.1 Pro | 80.2% | 76.8% | estimated ± 0.9 pp, low confidence |
| Gemini 3.5 Flash | 84.2% | 76.8% | estimated ± 0.9 pp, low confidence |
| Gemini 3.7 Flash | 88.7% | 85.5% | estimated ± 0.9 pp, low confidence |
| GLM-5.3-Flash | 89.4% | 90.4% | estimated ± 0.9 pp, low confidence |
| GPT-5.4 | 82.8% | 76.8% | estimated ± 0.9 pp, low confidence |
| Grok 4.20 | 60.9% | 76.8% | estimated ± 0.9 pp, low confidence |
| Inkling | 82.0% | 76.8% | estimated ± 0.9 pp, low confidence |
| Inkling-Small | 81.3% | 76.8% | estimated ± 0.9 pp, low confidence |
| MiMo-V2.5 | 81.0% | 76.8% | estimated ± 0.9 pp, low confidence |
| Muse Glimmer 30B | 78.8% | 76.8% | estimated ± 0.9 pp, low confidence |
| Muse Spark | 86.4% | 77.4% | estimated ± 0.9 pp, low confidence |
| Muse Spark 1.1 | 88.4% | 83.5% | estimated ± 0.9 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 76.3% | 76.8% | estimated ± 0.9 pp, low confidence |
| Qwen3.6-27B | 78.4% | 76.8% | estimated ± 0.9 pp, low confidence |
| Qwen3.6-35B-A3B | 78.0% | 76.8% | estimated ± 0.9 pp, low confidence |
| Sakana Fugu | 85.1% | 76.9% | estimated ± 0.9 pp, low confidence |
| Sakana Fugu-Ultra | 86.6% | 77.6% | estimated ± 0.9 pp, low confidence |