Calibration
HMMT Feb 2026 → FrontierMath v2 (Tiers 1-3)
FrontierMath v2 (Tiers 1-3) is estimated from HMMT Feb 2026 with a Hill curve fitted on 11 models measured on both: y = 0.0000 + (1.2000 − 0.0000)·x^6.00 / (0.97583^6.00 + x^6.00), R² = 0.66, cross-validated error 9.0 pp. It is used for 2 estimates.
| Estimated model | HMMT Feb 2026 | FrontierMath v2 (Tiers 1-3) | Source |
|---|---|---|---|
| Inkling 2 Small | 95.8% | 56.7% | estimated ± 9.0 pp, medium confidence |
| DeepSeek V4 Flash 0731 | 94.8% | 54.8% | estimated ± 9.0 pp, medium confidence |
Fitted on
| Model | HMMT Feb 2026 | FrontierMath v2 (Tiers 1-3) |
|---|---|---|
| Claude Opus 4.5 | 85.3% | 34.4% |
| DeepSeek V4 Pro | 95.2% | 64.6% |
| GLM-5.1 | 82.6% | 36.8% |
| GLM-5.2 | 92.5% | 59.2% |
| Inkling-Small | 90.2% | 46.3% |
| Kimi K2.6 | 92.7% | 57.2% |
| Qwen3.6-27B | 84.3% | 35.1% |
| Qwen3.6-35B-A3B | 83.6% | 20.4% |
| Qwen3.6 Plus | 87.8% | 38.2% |
| Qwen3.7 Max | 97.1% | 64.6% |
| Qwen3.7 Plus | 92.9% | 34.4% |