Skip to content

Benchmark

MathArena USAMO 2025

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

Models measured
9
Scores recorded
9
Sources
1
IQ weight
1.22

Score distribution

10.4%
Mean
4.2%
Median
10.3%
Std dev
2.8%
10th percentile
24.4%
90th percentile
2.1%
Min
2.1%30.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.22 for chat models, measured over 9 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1DeepSeek R1· MathArena data30.1%
2Gemini 2.5 Pro· MathArena data24.4%
3o4 Minihigh· MathArena data19.1%
4Grok 3think· MathArena data4.8%
5Gemini 2.0 Flashthinking· MathArena data4.2%
6Claude Sonnet 3.7think· MathArena data3.7%
7QwQ 32B· MathArena data3%
8O1 Prohigh· MathArena data2.8%
9o3 Minihigh· MathArena data2.1%

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)