Skip to content

Benchmark

MathArena HMMT Nov 2025

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

Models measured
19
Scores recorded
20
Sources
1
IQ weight
0.98

Score distribution

90.3%
Mean
91.7%
Median
4.7%
Std dev
84.2%
10th percentile
94.2%
90th percentile
80%
Min
80%99.2%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.98 for chat models, measured over 19 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1GPT 5.2xhigh· MathArena data99.2%
2GPT 5.2high· MathArena data95.8%
3GLM 5· MathArena data94.2%
4Step 3.5 Flash· MathArena data94.2%
5DeepSeek V3.2 Speciale· MathArena data93.3%
6Gemini 3 Flashhigh· MathArena data93.3%
7Gemini 3 Pro Previewpreview· MathArena data93.3%
8Grok 4.1 Fastreasoning· MathArena data93.3%
9GLM 4.6· MathArena data91.7%
10GPT 5.1high· MathArena data91.7%
11DeepSeek V3.2think· MathArena data90%
12GPT OSS 120Bhigh· MathArena data90%
13GPT 5high· MathArena data89.2%
14Kimi K2 Thinking· MathArena data89.2%
15Kimi K2.5high· MathArena data89.2%
16Grok 4· MathArena data88.3%
17DeepSeek V3.2 Expthink· MathArena data84.2%
18GPT-5 minihigh· MathArena data84.2%
19GPT-5 nanohigh· MathArena data81.7%
20Gemini 2.5 Pro· MathArena data80%

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)