Skip to content

Benchmark

MathArena Kangaroo 2025 11-12

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

Models measured
21
Scores recorded
21
Sources
1
IQ weight
2.10

Score distribution

91.7%
Mean
93.3%
Median
6.7%
Std dev
81.7%
10th percentile
98.3%
90th percentile
76.7%
Min
76.7%100%
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 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
1Claude Opus 4.8max· MathArena data100%
2GPT 5.4xhigh· MathArena data98.3%
3GPT 5.5xhigh· MathArena data98.3%
4Kimi K3max· MathArena data98.3%
5Gemini 3 Pro Previewpreview· MathArena data97.5%
6Gemini 3.1 Pro Previewhigh· MathArena data97.5%
7Gemini 3.6 Flash· MathArena data96.7%
8Gemini 3 Flashhigh· MathArena data95.8%
9GPT 5.2high· MathArena data95.8%
10GPT 5high· MathArena data95%
11Gemini 3.5 Flashhigh· MathArena data93.3%
12GPT 5.1high· MathArena data91.7%
13Qwen3.5 397B A17B· MathArena data91.7%
14Kimi K2.5high· MathArena data90%
15Gemini 2.5 Pro· MathArena data89.2%
16Claude Opus 4.6high· MathArena data87.5%
17Claude Sonnet 4.5think· MathArena data87.5%
18GPT-5 minihigh· MathArena data85%
19Grok 4.1 Fastreasoning· MathArena data81.7%
20GLM 4.5V· MathArena data78.3%
21Grok 4· MathArena data76.7%

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)