Skip to content

Benchmark

MathArena 03/2026

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

Models measured
18
Scores recorded
19
Sources
1
IQ weight
1.24

Score distribution

13.5%
Mean
14.6%
Median
7.6%
Std dev
5.8%
10th percentile
19.2%
90th percentile
0%
Min
0%35.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.24 for chat models, measured over 18 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1Claude Opus 4.8max· MathArena data35.7%
2DeepSeek V4 Flashmax· MathArena data19.6%
3GPT-5.6 Solmax, via Codex· MathArena data19.2%
4GPT 5.4xhigh· MathArena data17.1%
5GPT 5.5xhigh· MathArena data17.1%
6leanstral-1-5· MathArena data17.1%
7Gemini 3.5 Flashhigh· MathArena data16.1%
8Kimi K2.6high· MathArena data15.6%
9DeepSeek V4 Promax· MathArena data15.2%
10Claude Opus 4.7high· MathArena data14.6%
11Gemini 3.1 Pro Previewhigh· MathArena data14.6%
12GLM 5· MathArena data12.1%
13GLM 5.2max· MathArena data12.1%
14Step 3.7 Flashhigh· MathArena data10.7%
15Qwen3.5 2B· MathArena data6.7%
16Claude Opus 4.6high· MathArena data5.8%
17Claude Opus 4.7xhigh· MathArena data5.8%
18GLM 5.1· MathArena data2.4%
19Step 3.5 Flash· MathArena data0%

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)