Calibration
FrontierMath v2 (Tier 4) → HMMT Feb 2026
HMMT Feb 2026 is estimated from FrontierMath v2 (Tier 4) with a linear curve fitted on 6 models measured on both: y = 0.3758·x + 0.8353, R² = 0.91, cross-validated error 1.5 pp. It is used for 33 estimates.
| Estimated model | FrontierMath v2 (Tier 4) | HMMT Feb 2026 | Source |
|---|---|---|---|
| Claude Haiku 5.5 | 46.3% | 100.0% | estimated ± 1.5 pp, low confidence |
| Claude Opus 4.8 | 56.1% | 100.0% | estimated ± 1.5 pp, low confidence |
| GPT-5.4 | 49.0% | 100.0% | estimated ± 1.5 pp, low confidence |
| GPT-5.4 Pro | 58.5% | 100.0% | estimated ± 1.5 pp, low confidence |
| GPT-5.6 Luna | 58.5% | 100.0% | estimated ± 1.5 pp, low confidence |
| GPT-6 Luna | 56.1% | 100.0% | estimated ± 1.5 pp, low confidence |
| Muse Spark 1.3 | 46.3% | 100.0% | estimated ± 1.5 pp, low confidence |
| Qwen3.8 Max | 46.3% | 100.0% | estimated ± 1.5 pp, low confidence |
| Gemini 3.7 Flash | 36.6% | 97.3% | estimated ± 1.5 pp, low confidence |
| Claude Opus 4.7 | 31.7% | 95.4% | estimated ± 1.5 pp, medium confidence |
| Grok 4.6 | 31.7% | 95.4% | estimated ± 1.5 pp, medium confidence |
| GPT-5.2 | 31.7% | 95.4% | estimated ± 1.5 pp, medium confidence |
| Claude Sonnet 5 | 29.3% | 94.5% | estimated ± 1.5 pp, medium confidence |
| GLM-5.3 | 29.3% | 94.5% | estimated ± 1.5 pp, medium confidence |
| Claude Opus 4.6 | 26.8% | 93.6% | estimated ± 1.5 pp, medium confidence |
| DeepSeek V4.1 Flash 1017 | 26.8% | 93.6% | estimated ± 1.5 pp, medium confidence |
| Gemini 3.1 Pro | 26.8% | 93.6% | estimated ± 1.5 pp, medium confidence |
| Gemini 3.5 Flash | 26.8% | 93.6% | estimated ± 1.5 pp, medium confidence |
| Grok 4.5 | 24.4% | 92.7% | estimated ± 1.5 pp, medium confidence |
| Gemini 3.6 Flash | 22.0% | 91.8% | estimated ± 1.5 pp, medium confidence |
| Gemini 3.8 Flash | 22.0% | 91.8% | estimated ± 1.5 pp, medium confidence |
| Gemini 3 Flash | 17.1% | 89.9% | estimated ± 1.5 pp, medium confidence |
| GLM-5.3-Flash | 17.1% | 89.9% | estimated ± 1.5 pp, medium confidence |
| Grok 4.7 | 17.1% | 89.9% | estimated ± 1.5 pp, medium confidence |
| Grok 4.3 | 14.6% | 89.0% | estimated ± 1.5 pp, medium confidence |
| GPT-5.4 nano | 12.2% | 88.1% | estimated ± 1.5 pp, medium confidence |
| Kimi K2.7 Code | 12.2% | 88.1% | estimated ± 1.5 pp, medium confidence |
| GPT-5.4 mini | 9.8% | 87.2% | estimated ± 1.5 pp, medium confidence |
| Inkling | 4.9% | 85.4% | estimated ± 1.5 pp, medium confidence |
| o4-mini | 4.9% | 85.4% | estimated ± 1.5 pp, medium confidence |
| Claude Sonnet 4.5 | 2.4% | 84.5% | estimated ± 1.5 pp, low confidence |
| Gemini 2.5 Pro | 0.0% | 83.5% | estimated ± 1.5 pp, low confidence |
| Gemini 3.5 Flash-Lite | 0.0% | 83.5% | estimated ± 1.5 pp, low confidence |
Fitted on
| Model | FrontierMath v2 (Tier 4) | HMMT Feb 2026 |
|---|---|---|
| Claude Opus 4.5 | 4.9% | 85.3% |
| DeepSeek V4 Pro | 26.8% | 95.2% |
| GLM-5.2 | 29.3% | 92.5% |
| Inkling-Small | 17.1% | 90.2% |
| Kimi K2.6 | 25.6% | 92.7% |
| Qwen3.7 Max | 34.1% | 97.1% |