Calibration
CharXiv → MedXpertQA (MM)
MedXpertQA (MM) is estimated from CharXiv with a linear curve fitted on 6 models measured on both: y = 0.4023·x + 0.4283, R² = 0.54, cross-validated error 6.4 pp. It is used for 13 estimates.
| Estimated model | CharXiv | MedXpertQA (MM) | Source |
|---|---|---|---|
| Claude Mythos 5 | 93.5% | 80.4% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.7 (Adaptive) | 91.0% | 79.4% | estimated ± 6.4 pp, low confidence |
| Claude Opus 4.8 | 89.9% | 79.0% | estimated ± 6.4 pp, low confidence |
| Claude Sonnet 4.6 | 77.4% | 74.0% | estimated ± 6.4 pp, low confidence |
| Claude Sonnet 5 | 88.3% | 78.4% | estimated ± 6.4 pp, low confidence |
| Gemini 3.1 Flash-Lite | 73.2% | 72.3% | estimated ± 6.4 pp, low confidence |
| Gemini 3.7 Flash | 88.7% | 78.5% | estimated ± 6.4 pp, low confidence |
| GLM-5.3-Flash | 89.4% | 78.8% | estimated ± 6.4 pp, low confidence |
| Muse Spark 1.1 | 88.4% | 78.4% | estimated ± 6.4 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 76.3% | 73.5% | estimated ± 6.4 pp, low confidence |
| Qwen3.5-122B-A10B | 77.2% | 73.9% | estimated ± 6.4 pp, low confidence |
| Sakana Fugu | 85.1% | 77.1% | estimated ± 6.4 pp, low confidence |
| Sakana Fugu-Ultra | 86.6% | 77.7% | estimated ± 6.4 pp, low confidence |