Calibration
MathVision → V*
V* is estimated from MathVision with a Hill curve fitted on 9 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.71017^6.00 + x^6.00), R² = 0.79, cross-validated error 5.1 pp. It is used for 10 estimates.
| Estimated model | MathVision | V* | Source |
|---|---|---|---|
| dots3-note Preview | 87.7% | 93.6% | estimated ± 5.1 pp, medium confidence |
| Gemma 4 12B | 79.7% | 80.0% | estimated ± 5.1 pp, medium confidence |
| Kimi K3 | 94.3% | 100.0% | estimated ± 5.1 pp, low confidence |
| Qwen3.7 Plus | 90.3% | 97.0% | estimated ± 5.1 pp, low confidence |
| Qwen3.8-27B | 90.0% | 96.7% | estimated ± 5.1 pp, low confidence |
| Qwen3.8-Flash-Next | 90.6% | 97.4% | estimated ± 5.1 pp, low confidence |
| Qwen3.8 Max | 95.2% | 100.0% | estimated ± 5.1 pp, low confidence |
| Qwen3.8-Omni-Flash | 91.8% | 98.8% | estimated ± 5.1 pp, low confidence |
| Seed 2.1 Pro | 92.6% | 99.7% | estimated ± 5.1 pp, low confidence |
| Seed 2.1 Turbo | 90.1% | 96.8% | estimated ± 5.1 pp, low confidence |