Skip to content

Benchmark

MathArena Kangaroo 2025 7-8

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

Models measured
20
Scores recorded
20
Sources
1
IQ weight
2.10

Score distribution

88.3%
Mean
89.2%
Median
6%
Std dev
80%
10th percentile
95.8%
90th percentile
75%
Min
75%96.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.10 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
1Claude Opus 4.8max· MathArena data96.7%
2Kimi K3max· MathArena data96.7%
3GPT 5.4xhigh· MathArena data95.8%
4GPT 5.5xhigh· MathArena data95.8%
5Gemini 3.5 Flashhigh· MathArena data93.3%
6Gemini 3 Pro Previewpreview· MathArena data91.7%
7GPT 5high· MathArena data90.8%
8Gemini 3.1 Pro Previewhigh· MathArena data90%
9Gemini 3.6 Flash· MathArena data90%
10Gemini 3 Flashhigh· MathArena data89.2%
11GPT 5.2high· MathArena data89.2%
12Kimi K2.5high· MathArena data88.3%
13GPT-5 minihigh· MathArena data87.5%
14Claude Opus 4.6high· MathArena data86.7%
15GPT 5.1high· MathArena data85.8%
16Gemini 2.5 Pro· MathArena data82.5%
17Grok 4· MathArena data80.8%
18Claude Sonnet 4.5think· MathArena data80%
19Grok 4.1 Fastreasoning· MathArena data79.2%
20GLM 4.5V· MathArena data75%

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)