Atlas

Benchmarks

← All benchmarks

SWE Atlas - Refactoring

Code · 2026-03-04

Score on SWE Atlas refactoring tasks that restructure code while preserving behavior.

Top models (higher is better)

ModelScore
Claude Fable 554.8
Opus 4.748.6
Opus 4.846.7
GPT-5.544.8
GPT-5.444.3
GLM-5.242.4
GPT-5.3-Codex42.4
Opus 4.635.6
Gemini 3.1 Pro Preview33.8
Sonnet 4.632.2
GLM-524.2
Kimi K2.520.9
MiniMax M2.519.5
Gemini 3 Flash Preview10.0
Loading Atlas data…