Skip to content

Benchmark

MathArena 10/2026

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

Models measured
11
Scores recorded
11
Sources
1
IQ weight
1.21

Score distribution

64.4%
Mean
73.5%
Median
21.9%
Std dev
33.3%
10th percentile
88.9%
90th percentile
30.6%
Min
30.6%93.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.21 for chat models, measured over 11 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-6.2 Astramax· MathArena data93.8%
2GPT-6.2 Solmax· MathArena data88.9%
3Gemini 4 Carbonhigh· MathArena data82.7%
4GPT-6.1 Solmax· MathArena data80.3%
5Claude Opus 5.5high· MathArena data77.8%
6Claude Fable 5.2max· MathArena data73.5%
7Kimi K3.1 Promax· MathArena data58%
8DeepSeek V4.1 Flash 1017max, DSH· MathArena data49.7%
9Gemini 4 Argonhigh· MathArena data39.8%
10Muse Spark Openxhigh· MathArena data33.3%
11Mistral Large 4high· MathArena data30.6%

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)