Atlas

Benchmarks

← All benchmarks

USAMO 2026

Math · 2026-03-28

2026 USAMO proof-grading benchmark at med/high/xhigh effort.

Top models (higher is better)

ModelScore
Claude Mythos 599.8
GPT-5.598.2
Claude Mythos Preview97.6
Opus 4.896.7
GPT-5.495.2
Sonnet 579.5
Gemini 3.1 Pro Preview74.4
Opus 4.769.3
DeepSeek-V4-Pro60.7
Kimi K2.651.2
Opus 4.647.0
Step 3.5 Flash44.6
Qwen3.5 397B A17B36.3
GLM-535.1
Loading Atlas data…