Skip to content

Benchmark

MathArena Kangaroo 2025 3-4

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

Models measured
21
Scores recorded
21
Sources
1
IQ weight
1.89

Score distribution

65.6%
Mean
66.7%
Median
11.6%
Std dev
50%
10th percentile
77.1%
90th percentile
39.6%
Min
39.6%89.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.89 for chat models, measured over 21 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 5.5xhigh· MathArena data89.6%
2GPT 5.4xhigh· MathArena data83.3%
3Kimi K3max· MathArena data77.1%
4Gemini 3.1 Pro Previewhigh· MathArena data76%
5GPT 5.2high· MathArena data74%
6Gemini 3.5 Flashhigh· MathArena data72.9%
7Gemini 3.6 Flash· MathArena data68.8%
8Qwen3.5 397B A17B· MathArena data68.8%
9Gemini 3 Flashhigh· MathArena data66.7%
10Gemini 3 Pro Previewpreview· MathArena data66.7%
11GPT-5 minihigh· MathArena data66.7%
12GPT 5.1high· MathArena data65.6%
13Kimi K2.5high· MathArena data65.6%
14Gemini 2.5 Pro· MathArena data64.6%
15Claude Sonnet 4.5think· MathArena data62.5%
16Claude Opus 4.8max· MathArena data60.4%
17GPT 5high· MathArena data60.4%
18Grok 4· MathArena data52.1%
19Claude Opus 4.6high· MathArena data50%
20GLM 4.5V· MathArena data45.8%
21Grok 4.1 Fastreasoning· MathArena data39.6%

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)