Calibration
HMMT Feb 2026 → FrontierMath v2 (Tiers 1-3)
FrontierMath v2 (Tiers 1-3) is estimated from HMMT Feb 2026 with a offset logistic curve fitted on 6 models measured on both: y = 0.2438 + (0.3897 − 0.2438) / (1 + exp(−200.00·(x − 0.8871))), R² = 0.50, cross-validated error 8.4 pp. It is used for 17 estimates.
| Estimated model | HMMT Feb 2026 | FrontierMath v2 (Tiers 1-3) | Source |
|---|---|---|---|
| DeepSeek V4 Flash 0731 | 94.8% | 39.0% | estimated ± 8.4 pp, low confidence |
| DeepSeek V4 Pro 0813 | 95.2% | 39.0% | estimated ± 8.4 pp, low confidence |
| GLM-5.2 | 92.5% | 39.0% | estimated ± 8.4 pp, low confidence |
| Inkling-Small | 90.2% | 38.3% | estimated ± 8.4 pp, low confidence |
| K-EXAONE 2.0 | 78.4% | 24.4% | estimated ± 8.4 pp, low confidence |
| Ling 3.0 Flash | 87.0% | 24.8% | estimated ± 8.4 pp, low confidence |
| LongCat-Flash-Lite-Sparse | 40.5% | 24.4% | estimated ± 8.4 pp, low confidence |
| MAI-Thinking-1 | 84.9% | 24.4% | estimated ± 8.4 pp, low confidence |
| MiniCPM5-1B | 25.8% | 24.4% | estimated ± 8.4 pp, low confidence |
| MiniCPM5-2B | 63.8% | 24.4% | estimated ± 8.4 pp, low confidence |
| Qwen3.5 397B | 87.9% | 26.8% | estimated ± 8.4 pp, low confidence |
| Qwen3.6-27B | 84.3% | 24.4% | estimated ± 8.4 pp, low confidence |
| Qwen3.6-35B-A3B | 83.6% | 24.4% | estimated ± 8.4 pp, low confidence |
| Qwen3.7 Max | 97.1% | 39.0% | estimated ± 8.4 pp, low confidence |
| Qwen3.7 Plus | 92.9% | 39.0% | estimated ± 8.4 pp, low confidence |
| Solar Open 2 | 93.9% | 39.0% | estimated ± 8.4 pp, low confidence |
| ZAYA1-8B | 71.6% | 24.4% | estimated ± 8.4 pp, low confidence |