Skip to content

Benchmark

MathArena Kangaroo 2025 1-2

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

Models measured
21
Scores recorded
21
Sources
1
IQ weight
0.99

Score distribution

75.4%
Mean
76%
Median
12.3%
Std dev
61.5%
10th percentile
93.8%
90th percentile
59.4%
Min
59.4%95.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.99 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 data95.8%
2GPT 5.4xhigh· MathArena data94.8%
3Kimi K3max· MathArena data93.8%
4Gemini 3.5 Flashhigh· MathArena data89.6%
5Gemini 3 Flashhigh· MathArena data87.5%
6Gemini 3.1 Pro Previewhigh· MathArena data86.5%
7Gemini 3.6 Flash· MathArena data83.3%
8Qwen3.5 397B A17B· MathArena data81.3%
9GPT 5.2high· MathArena data80.2%
10Gemini 3 Pro Previewpreview· MathArena data76%
11Kimi K2.5high· MathArena data76%
12Claude Opus 4.8max· MathArena data70.8%
13GPT 5high· MathArena data68.8%
14GLM 4.5V· MathArena data65.6%
15GPT 5.1high· MathArena data65.6%
16Gemini 2.5 Pro· MathArena data64.6%
17Claude Sonnet 4.5think· MathArena data61.5%
18GPT-5 minihigh· MathArena data61.5%
19Grok 4· MathArena data61.5%
20Grok 4.1 Fastreasoning· MathArena data60.4%
21Claude Opus 4.6high· MathArena data59.4%

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)