Skip to content

Benchmark

MathArena Kangaroo 2025 5-6

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

Models measured
21
Scores recorded
21
Sources
1
IQ weight
1.13

Score distribution

73.8%
Mean
70.8%
Median
9.8%
Std dev
62.5%
10th percentile
88.3%
90th percentile
58.3%
Min
58.3%90%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.13 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
1Gemini 3.5 Flashhigh· MathArena data90%
2GPT 5.5xhigh· MathArena data90%
3Gemini 3.6 Flash· MathArena data88.3%
4Gemini 3.1 Pro Previewhigh· MathArena data86.7%
5GPT 5.4xhigh· MathArena data83.3%
6Kimi K3max· MathArena data81.7%
7GPT 5.2high· MathArena data80%
8Gemini 3 Flashhigh· MathArena data77.5%
9Gemini 3 Pro Previewpreview· MathArena data76.7%
10Qwen3.5 397B A17B· MathArena data74.9%
11GPT-5 minihigh· MathArena data70.8%
12Claude Opus 4.8max· MathArena data70%
13Claude Sonnet 4.5think· MathArena data68.3%
14Kimi K2.5high· MathArena data67.5%
15Gemini 2.5 Pro· MathArena data66.7%
16Grok 4.1 Fastreasoning· MathArena data65.8%
17GPT 5high· MathArena data65%
18Grok 4· MathArena data63.3%
19GLM 4.5V· MathArena data62.5%
20GPT 5.1high· MathArena data61.7%
21Claude Opus 4.6high· MathArena data58.3%

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)