Calibration
ERQA → MathVision
MathVision is estimated from ERQA with a Hill curve fitted on 7 models measured on both: y = 0.8518 + (1.2000 − 0.8518)·x^6.00 / (0.91995^6.00 + x^6.00), R² = 0.74, cross-validated error 1.4 pp. It is used for 6 estimates.
| Estimated model | ERQA | MathVision | Source |
|---|---|---|---|
| Claude Opus 4.6 | 51.6% | 86.2% | estimated ± 1.4 pp, low confidence |
| Gemini 3.1 Pro | 69.4% | 90.6% | estimated ± 1.4 pp, medium confidence |
| GPT-5.4 | 65.4% | 89.2% | estimated ± 1.4 pp, low confidence |
| Grok 4.20 | 54.1% | 86.6% | estimated ± 1.4 pp, low confidence |
| Muse Spark | 64.7% | 88.9% | estimated ± 1.4 pp, low confidence |
| Qwen3.6-27B | 62.5% | 88.3% | estimated ± 1.4 pp, low confidence |