HMMT February 2025

Name: HMMT February 2025
Published: 2025-02-15

Math · 2025-02-15

The 30 problems of the February 2025 Harvard-MIT Mathematics Tournament, scored by final-answer correctness as run by a model developer with its own sampling, usually averaging several samples per problem. Shares a factor group with the MathArena track on the same problem set.

Top models (higher is better)

Model	Score
GPT-5.2	99.4
Qwen3 Max Thinking (2026-01-23)	98.0
Gemini 3 Pro Preview	97.5
DeepSeek-V3.2	92.5
Qwen3 235B A22B Thinking 2507	83.9
Gemini 2.5 Pro	82.5
DeepSeek-R1-0528	79.4
o3	77.5
Qwen3 VL 235B A22B Thinking	77.4
Qwen3-Next-80B-A3B-Thinking	73.9
Qwen3 30B A3B Thinking 2507	71.4
Qwen3-Coder-Next	70.2
Qwen3 VL 30B A3B Thinking	67.6
O4 Mini	66.7
Qwen3 VL 32B Thinking	64.6
Gemini 2.5 Flash	64.2
Qwen3-235B-A22B	62.5
Qwen3 VL 8B Thinking	60.6
Opus 4	58.3
Qwen3-VL-235B-A22B-Instruct	57.4
Qwen3 4B Thinking 2507	55.5
Qwen3 235B A22B Instruct 2507	55.4
Qwen3 Next 80B A3B Instruct	54.1
Qwen3 VL 4B Thinking	53.1
Qwen3 VL 30B A3B Instruct	50.6
Qwen3-30B-A3B	49.8
Qwen3 VL 32B Instruct	46.1
Qwen3 8B	43.2
Qwen3 30B A3B Instruct 2507	43.0
Qwen3 4B	42.1
Kimi K2 Instruct	38.8
Qwen3-VL-8B-Instruct	32.5
Qwen3 4B Instruct 2507	31.0
Qwen3 VL 4B Instruct	30.7
DeepSeek-V3-0324	27.5
Qwen3 1.7B	24.3
Qwen3 VL 2B Thinking	22.8
Qwen3 VL 2B Instruct	10.9
GPT-4.1 Nano	9.7
ChatGPT-4o Latest observed 2025-03-27	7.9