benchgap
Math

HMMT Feb 2025 leaderboard

As of 2026-11-22, the highest measured score on HMMT Feb 2025 is 97.5% by GLM-5. 44 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Beam 1.0 501B-A23B FP898.6%estimated ± 1.9 pp, low confidence
2Beam 1.0 501B-A23B98.3%estimated ± 1.9 pp, low confidence
3Beam 1.0 501B-A23B NVFP498.0%estimated ± 1.9 pp, low confidence
4A.X K297.6%estimated ± 1.9 pp, low confidence
5Inkling97.6%estimated ± 1.9 pp, low confidence
6GLM-597.5%measured
7GLM-4.797.1%measured
8Ripple-65B Preview97.1%estimated ± 1.8 pp, low confidence
9Inkling-Small96.9%estimated ± 1.8 pp, low confidence
10Kimi K2.696.8%estimated ± 1.9 pp, low confidence
11Qwen3.6 Plus96.7%measured
12GPT-5.296.6%estimated ± 2.9 pp, medium confidence
13Step 3.5 Flash96.5%estimated ± 2.9 pp, medium confidence
14OrcaCyber-Nano-1.096.5%estimated ± 2.9 pp, medium confidence
15GLM-5.296.3%estimated ± 1.9 pp, high confidence
16MiMo-V2-Flash96.2%estimated ± 2.9 pp, high confidence
17Ternary Bonsai 2 27B96.1%estimated ± 1.9 pp, low confidence
18Solar Open 296.0%estimated ± 1.9 pp, medium confidence
19o4-mini95.6%estimated ± 2.9 pp, high confidence
20GPT-OSS 120B95.5%estimated ± 2.9 pp, high confidence
21Solar Pro 495.5%estimated ± 1.9 pp, medium confidence
22Kimi K2.595.4%measured
23Ling 3.0 Flash95.1%estimated ± 1.8 pp, low confidence
24GPT-OSS 20B94.9%estimated ± 2.9 pp, high confidence
25Qwen3.5 397B94.8%measured
26GLM-4.7-Flash94.8%estimated ± 2.9 pp, high confidence
27Muse Glimmer 30B94.7%estimated ± 1.9 pp, medium confidence
28MAI-Thinking-193.9%estimated ± 1.8 pp, low confidence
29Qwen3.6-27B93.8%measured
30Nemotron 3 Super 100B93.7%measured
31Claude Opus 4.592.9%measured
32GLM-5.192.5%estimated ± 1.8 pp, low confidence
33Spark-X2.6-4B91.7%estimated ± 1.8 pp, low confidence
34K-EXAONE 2.091.5%estimated ± 1.9 pp, low confidence
35Qwen3.6-35B-A3B90.7%measured
36Nemotron 3 Nano 30B89.4%estimated ± 2.9 pp, high confidence
37Granite 4.2 30B89.2%measured
38Sarvam 105B85.8%measured
39DeepSeek-R1-052882.4%estimated ± 2.9 pp, high confidence
40MiniMax M2.781.4%estimated ± 1.9 pp, high confidence
41Claude Sonnet 4.579.8%estimated ± 2.9 pp, high confidence
42Granite 4.2 8B78.3%measured
43MiniCPM5-2B77.3%estimated ± 2.9 pp, high confidence
44Sarvam 30B73.3%measured
45Exaone 4.0 32B71.9%estimated ± 2.9 pp, high confidence
46Granite 4.2 3B66.7%measured
47Nemotron 3 Nano Omni 30B A3B65.3%estimated ± 2.9 pp, high confidence
48Qwen3 235B-A22B (Reasoning)64.9%estimated ± 2.9 pp, high confidence
49Phi-4-reasoning-plus64.1%estimated ± 2.9 pp, high confidence
50MiniMax M1 80k64.1%estimated ± 2.9 pp, high confidence
51Claude 4 Opus Thinking64.0%estimated ± 2.9 pp, medium confidence
52Qwen3 32B (Reasoning)64.0%estimated ± 2.9 pp, medium confidence
53Nemotron Ultra 253B64.0%estimated ± 2.9 pp, medium confidence
54Qwen3 235B 250764.0%estimated ± 2.9 pp, medium confidence
55QwQ-32B64.0%estimated ± 2.9 pp, medium confidence
56Magistral Medium64.0%estimated ± 2.9 pp, medium confidence
57Phi-4-reasoning64.0%estimated ± 2.9 pp, medium confidence
58Magistral Small64.0%estimated ± 2.9 pp, medium confidence
59DeepSeek-R1-0528-Qwen3-8B61.5%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General