Skip to content

Benchmark

USAMO 2026

Math benchmarks measure how well models solve maths problems. Scores are percentages, except where a source reports on a different scale. Check each score's scale before comparing. Higher is better.

Models measured
4
Scores recorded
6
Sources
2
IQ weight
No weight published

Score distribution

53.4%
Mean
69.3%
Median
20.5%
Std dev
33.4%
10th percentile
79.5%
90th percentile
30.2%
Min

Leaderboard

#ModelScore
1MiniMax M3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim36
2Claude Sonnet 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim33.4
3Claude Sonnet 5high effort · 300K token limit · 10 attempts per problem· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.79.5%
4Hy3 (Deep Infra)· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim30.2
5Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72%
6Claude Opus 4.7MathArena grading · medium effort · 300K token limit · avg@10 per problem· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.69.3%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats