Skip to content

Benchmark

MathArena CMIMC 2025

Math benchmarks measure how well models solve maths problems. Scores are percentages.

Models measured
29
Scores recorded
32
Sources
1
IQ weight
1.06

Score distribution

78.8%
Mean
84.4%
Median
14.2%
Std dev
58.1%
10th percentile
94.4%
90th percentile
40%
Min
40%95.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.06 for chat models, measured over 29 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1DeepSeek V3.2 Speciale· MathArena data95.6%
2GPT 5.2high· MathArena data95.6%
3Step 3.5 Flash· MathArena data95.6%
4GPT 5.1high· MathArena data94.4%
5GLM 5· MathArena data93.1%
6Kimi K2 Thinking· MathArena data90.6%
7Kimi K2.5high· MathArena data90.6%
8Gemini 3 Flashhigh· MathArena data90%
9Gemini 3 Pro Previewpreview· MathArena data90%
10GPT 5high· MathArena data88.8%
11GLM 4.6· MathArena data87.5%
12GPT OSS 120Bhigh· MathArena data85%
13Grok 4· MathArena data85%
14DeepSeek V3.2think· MathArena data84.4%
15GPT-5 minihigh· MathArena data84.4%
16Grok 4.1 Fastreasoning· MathArena data84.4%
17o4 Minihigh· MathArena data83.8%
18DeepSeek V3.1think· MathArena data82.5%
19OpenAI o3high· MathArena data82.5%
20DeepSeek V3.2 Expthink· MathArena data76.3%
21GPT-5 nanohigh· MathArena data74.4%
22GPT OSS 20bhigh· MathArena data73.8%
23GLM 4.5· MathArena data72.5%
24GLM 4.5 Air· MathArena data71.9%
25DeepSeek R1· MathArena data71.3%
26Grok 3 Minihigh· MathArena data68.1%
27Claude Sonnet 4.5think· MathArena data66.3%
28o4 Minimedium· MathArena data62.5%
29Gemini 2.5 Pro· MathArena data58.1%
30Gemini 2.5 Flashthinking· MathArena data54.4%
31o4 Minilow· MathArena data48.1%
32Grok 3 Minilow· MathArena data40%

Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.

Data credits: MathArena (ETH SRI / INSAIT), MIT — cite dekoninck2026matharena (licence)