Atlas

Benchmarks

← All benchmarks

BrokenArXiv 05/2026

Math · 2026-06-11

This MathArena track evaluates models on the BrokenArXiv 05/2026 problem set. Scores report the percentage of problems answered correctly.

Top models (higher is better)

ModelScore
GPT-5.550.0
Claude Fable 544.5
Kimi K342.5
Opus 4.823.0
DeepSeek-V4-Flash19.5
Gemini 3.1 Pro Preview17.0
Gemini 3.5 Flash11.0
GLM-5.210.5
GLM-5.110.0
Step 3.7 Flash8.5
Gemini 3.6 Flash4.0
Loading Atlas data…