Skip to content

Benchmark

MathArena Kangaroo 2025 9-10

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

Models measured
20
Scores recorded
20
Sources
1
IQ weight
1.58

Score distribution

94.8%
Mean
96.7%
Median
5.7%
Std dev
87.5%
10th percentile
100%
90th percentile
78.3%
Min
78.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.58 for chat models, measured over 20 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1Gemini 3.1 Pro Previewhigh· MathArena data100%
2Gemini 3.5 Flashhigh· MathArena data100%
3Gemini 3.6 Flash· MathArena data100%
4GPT 5.2high· MathArena data100%
5GPT 5.5xhigh· MathArena data100%
6Kimi K3max· MathArena data100%
7GPT 5.4xhigh· MathArena data99.2%
8Gemini 3 Flashhigh· MathArena data98.3%
9GPT-5 minihigh· MathArena data97.5%
10Gemini 3 Pro Previewpreview· MathArena data96.7%
11Gemini 2.5 Pro· MathArena data95.8%
12Kimi K2.5high· MathArena data95.8%
13Claude Sonnet 4.5think· MathArena data95%
14GPT 5high· MathArena data92.5%
15Claude Opus 4.6high· MathArena data91.7%
16Claude Opus 4.8max· MathArena data91.7%
17GPT 5.1high· MathArena data90.8%
18Grok 4.1 Fastreasoning· MathArena data87.5%
19Grok 4· MathArena data85.8%
20GLM 4.5V· MathArena data78.3%

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)