Atlas

Benchmarks

← All benchmarks

ArXivMath — Overall

Math · 2026-02-04

This MathArena aggregate covers the full ArXivMath problem collection. Scores report the percentage of problems answered correctly.

Top models (higher is better)

ModelScore
Claude Fable 580.6
GPT-5.575.6
Opus 4.864.8
Kimi K362.5
Gemini 3.1 Pro Preview60.0
Gemini 3.6 Flash54.2
GLM-5.250.5
Gemini 3.5 Flash49.5
DeepSeek-V4-Flash49.2
Step 3.7 Flash38.0
Qwen3.6 35B A3B32.7
Qwen3.5-2B5.7
Loading Atlas data…