Benchmarks / Math
HMMT February 2026 (MathArena)
The 33 final-answer problems of the February 2026 Harvard-MIT Mathematics Tournament, one of the largest high-school math competitions in the United States, run by MathArena on the contest problems. Maintained by MathArena. Score: Accuracy (average performance on the competition) with a 95% confidence interval (normal approximation).
HMMT February 2026 (MathArena)
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-5.5 | xhigh | 98.5% |
| 2 | GPT-5.4 | xhigh | 97.7% |
| 3 | GPT-5.2 | high | 97.0% |
| 4 | Kimi K3 | max | 97.0% |
| 5 | Opus 4.6 | high | 96.2% |
| 6 | Gemini 3.5 Flash | xhigh | 95.5% |
| 7 | Opus 4.8 | max | 95.5% |
| 8 | Kimi K2.6 | max | 94.7% |
| 9 | Gemini 3.1 Pro | high | 94.7% |
| 10 | DeepSeek V4 Flash | max | 93.9% |
| 11 | DeepSeek V4 Pro | max | 93.9% |
| 12 | Opus 4.7 | xhigh | 93.9% |
| 13 | GLM 5.2 | max | 92.4% |
| 14 | Gemini 3 Flash | high | 89.4% |
| 15 | Gemini 3.6 Flash | xhigh | 89.4% |
| 16 | GLM 5.1 | max | 89.4% |
| 17 | Qwen 3.5 397B-A17B | max | 87.9% |
| 18 | Kimi K2.5 | max | 87.1% |
| 19 | Grok 4.1 Fast | high | 86.4% |
| 20 | GLM 5 | high | 86.4% |
| 21 | Gemini 3 Pro | high | 86.4% |
| 22 | DeepSeek V3.2 | high | 84.1% |
| 23 | Qwen 3.5 35B-A3B | low | 81.8% |
| 24 | Qwen 3.5 27B | high | 81.1% |
| 25 | Qwen 3 30B-A3B | none | 78.8% |
| 26 | Qwen 3.5 9B | low | 71.2% |
| 27 | Qwen 3 4B | none | 53.0% |