Calibration
HMMT Feb 2026 → MMAnswerBench
MMAnswerBench is estimated from HMMT Feb 2026 with a Michaelis–Menten + offset curve fitted on 10 models measured on both: y = 0.0000 + 2.0000·x / (1.21710 + x), R² = 0.49, cross-validated error 2.8 pp. It is used for 4 estimates.
| Estimated model | HMMT Feb 2026 | MMAnswerBench | Source |
|---|---|---|---|
| DeepSeek V4 Flash 0731 | 94.8% | 87.6% | estimated ± 2.8 pp, low confidence |
| DeepSeek V4 Pro 0813 | 95.2% | 87.8% | estimated ± 2.8 pp, low confidence |
| Qwen3.7 Max | 97.1% | 88.8% | estimated ± 2.8 pp, low confidence |
| Qwen3.7 Plus | 92.9% | 86.6% | estimated ± 2.8 pp, low confidence |