benchgap
Math

MATH-500 leaderboard

As of 2026-11-21, the highest measured score on MATH-500 is 98.8% by Ternary Bonsai 2 27B. 63 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-5.299.4%estimated ± 2.0 pp, medium confidence
2Step 3.5 Flash99.1%estimated ± 2.0 pp, medium confidence
3MAI-Thinking-199.1%estimated ± 2.0 pp, medium confidence
4Kimi K2.599.0%estimated ± 2.0 pp, high confidence
5GLM-4.798.9%estimated ± 2.0 pp, high confidence
6Ternary Bonsai 2 27B98.8%measured
7MiMo-V2-Flash98.7%estimated ± 2.0 pp, high confidence
8OrcaCyber-Nano-1.098.6%measured
9Sarvam 105B98.6%measured
10GLM-4.7-Flash98.4%estimated ± 2.0 pp, high confidence
11Nemotron 3 Super 100B98.3%estimated ± 2.0 pp, high confidence
12Ripple-65B Preview98.3%estimated ± 1.2 pp, medium confidence
13GLM-5.298.2%estimated ± 3.1 pp, low confidence
14GPT-OSS 120B98.2%estimated ± 1.2 pp, medium confidence
15Granite 4.2 30B98.1%estimated ± 2.0 pp, high confidence
16Nemotron 3 Nano 30B98.1%estimated ± 2.0 pp, high confidence
17Beam 1.0 501B-A23B FP898.0%estimated ± 3.1 pp, low confidence
18Beam 1.0 501B-A23B98.0%estimated ± 3.1 pp, low confidence
19Qwen3 235B-A22B (Reasoning)98.0%measured
20QwQ-32B98.0%measured
21Beam 1.0 501B-A23B NVFP498.0%estimated ± 3.1 pp, low confidence
22GPT-OSS 20B97.9%estimated ± 1.2 pp, medium confidence
23A.X K297.9%estimated ± 3.1 pp, low confidence
24Inkling97.9%estimated ± 3.1 pp, low confidence
25DeepSeek-R1-052897.9%estimated ± 1.2 pp, medium confidence
26Claude Sonnet 4.597.9%estimated ± 2.0 pp, high confidence
27Granite 4.2 8B97.8%estimated ± 2.0 pp, high confidence
28Kimi K2.697.8%estimated ± 3.1 pp, low confidence
29GLM-597.7%estimated ± 3.1 pp, medium confidence
30Solar Open 297.7%estimated ± 3.1 pp, medium confidence
31Gemini 2.5 Pro97.7%estimated ± 1.3 pp, medium confidence
32Inkling-Small97.7%estimated ± 3.1 pp, medium confidence
33GLM-5.197.7%estimated ± 3.1 pp, medium confidence
34Qwen3.6 Plus97.7%estimated ± 3.1 pp, medium confidence
35Solar Pro 497.7%estimated ± 3.1 pp, medium confidence
36Exaone 4.0 32B97.7%estimated ± 2.0 pp, high confidence
37Claude Opus 4.597.7%estimated ± 3.1 pp, medium confidence
38Muse Glimmer 30B97.6%estimated ± 3.1 pp, medium confidence
39Grok 3 Mini97.5%estimated ± 1.3 pp, medium confidence
40Qwen3.6-27B97.5%estimated ± 3.1 pp, medium confidence
41o4-mini97.5%measured
42DeepSeek-R1-0528-Qwen3-8B97.5%estimated ± 1.2 pp, high confidence
43Qwen3.5 397B97.4%estimated ± 3.1 pp, medium confidence
44Kimi K297.4%measured
45o3-mini97.4%estimated ± 1.3 pp, medium confidence
46Ling 3.0 Flash97.4%estimated ± 3.1 pp, medium confidence
47Qwen3.6-35B-A3B97.3%estimated ± 3.1 pp, medium confidence
48DeepSeek-R197.3%measured
49Nemotron 3 Nano Omni 30B A3B97.3%estimated ± 2.0 pp, high confidence
50K-EXAONE 2.097.3%estimated ± 3.1 pp, medium confidence
51Qwen3 32B (Reasoning)97.2%measured
52Spark-X2.6-4B97.1%estimated ± 3.1 pp, medium confidence
53Phi-4-reasoning-plus97.1%estimated ± 1.2 pp, high confidence
54Nemotron Ultra 253B97.0%measured
55Sarvam 30B97.0%measured
56MiniMax M2.796.9%estimated ± 3.1 pp, medium confidence
57Granite 4.2 3B96.8%estimated ± 2.0 pp, high confidence
58ZAYA1-8B96.8%estimated ± 3.1 pp, medium confidence
59MiniMax M1 80k96.8%measured
60Phi-4-reasoning96.6%estimated ± 1.2 pp, high confidence
61Claude 4 Opus Thinking96.5%estimated ± 2.0 pp, high confidence
62Claude 3.7 Sonnet Thinking96.2%measured
63LFM2.5-24B-A2B96.0%measured
64Magistral Small95.9%measured
65Qwen3 235B 250795.9%estimated ± 2.0 pp, high confidence
66LongCat-Flash-Lite-Sparse95.8%measured
67DeepSeek V3.195.6%estimated ± 1.2 pp, high confidence
68Gemma 4 12B95.2%estimated ± 3.1 pp, medium confidence
69ZAYA1-74B-Preview95.1%estimated ± 3.1 pp, medium confidence
70MiniCPM5-2B94.6%measured
71DeepSeek R1 Distill Llama 70B94.5%measured
72Magistral Medium94.3%measured
73DeepSeek R1 Distill Qwen 14B93.9%measured
74LFM2.5-2.6B93.7%estimated ± 2.0 pp, high confidence
75Grok 3 [Beta]93.5%estimated ± 1.3 pp, high confidence
76Gemini 2.5 Flash-Lite93.4%estimated ± 2.0 pp, high confidence
77MiniCPM5-1B91.6%measured
78Claude 4 Opus91.5%estimated ± 2.0 pp, high confidence
79Claude 4 Sonnet91.4%estimated ± 2.0 pp, high confidence
80Qwen3 235B-A22B91.2%measured
81Gemini 2.0 Flash91.0%estimated ± 2.0 pp, high confidence
82Gemini 2.0 Flash-Lite90.3%estimated ± 2.0 pp, high confidence
83DeepSeek V390.2%measured
84o1-mini90.0%measured
85LLaDA2.2-mini89.5%estimated ± 3.1 pp, low confidence
86Gemini 1.5 Flash89.2%estimated ± 2.0 pp, medium confidence
87LFM2.5-8B-A1B88.8%measured
88Qwen3 32B88.6%measured
89Claude 3.7 Sonnet82.2%measured
90Kanana-2 1.3B Instruct61.4%measured
91Kanana-2 3B Instruct61.2%measured
92Claude 3.5 Haiku49.5%estimated ± 1.3 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General