Calibration
MathVision → BabyVision
BabyVision is estimated from MathVision with a linear curve fitted on 5 models measured on both: y = 4.1091·x + -3.0756, R² = 0.96, cross-validated error 4.3 pp. It is used for 14 estimates.
| Estimated model | MathVision | BabyVision | Source |
|---|---|---|---|
| Claude Opus 4.5 | 74.3% | 0.0% | estimated ± 4.3 pp, low confidence |
| Gemini 3 Pro | 86.6% | 48.3% | estimated ± 4.3 pp, low confidence |
| Gemma 4 12B | 79.7% | 19.9% | estimated ± 4.3 pp, low confidence |
| GPT-5.2 | 83.0% | 33.5% | estimated ± 4.3 pp, low confidence |
| Kimi K2.6 | 87.4% | 51.6% | estimated ± 4.3 pp, low confidence |
| Kimi K3 | 94.3% | 79.9% | estimated ± 4.3 pp, medium confidence |
| Qwen3.5-122B-A10B | 86.2% | 46.6% | estimated ± 4.3 pp, low confidence |
| Qwen3.5-27B | 86.0% | 45.8% | estimated ± 4.3 pp, low confidence |
| Qwen3.5-35B-A3B | 83.9% | 37.2% | estimated ± 4.3 pp, low confidence |
| Qwen3.5 397B | 88.6% | 56.5% | estimated ± 4.3 pp, medium confidence |
| Qwen3.6 Plus | 88.0% | 54.0% | estimated ± 4.3 pp, medium confidence |
| Qwen3.7 Plus | 90.3% | 63.5% | estimated ± 4.3 pp, medium confidence |
| Qwen3.8-Flash-Next | 90.6% | 64.7% | estimated ± 4.3 pp, medium confidence |
| Qwen3.8-Omni-Flash | 91.8% | 69.7% | estimated ± 4.3 pp, medium confidence |