Atlas

Benchmarks

← All benchmarks

CadQueryEval Corrected Geometry Components

Code · 2026-09-30

CadQueryEval final corrected-geometry release: percentage of 25 meshes matching the task's expected component count.

Top models (higher is better)

ModelScore
Claude Opus 5.5100.0
Gemini 3.1 Pro Preview100.0
Gemini 3.8 Flash100.0
GPT-5.6 Sol Pro100.0
GPT-6 Astra100.0
GPT-6 Luna100.0
GPT-6 Sol100.0
Gemini 3.7 Flash96.0
GPT-5.6 Sol96.0
Grok 4.796.0
GLM-5.392.0
GLM-5.3 Flash92.0
GPT-5.592.0
GPT-5.6 Luna Pro92.0
GPT-5.6 Terra Pro92.0
Claude Fable 5.188.0
Opus 4.888.0
Claude Opus 588.0
GPT-5.6 Luna88.0
GPT-5.6 Terra88.0
Grok 4.588.0
Kimi K388.0
MiMo-V2.6-Pro88.0
Muse Spark 1.388.0
Qwen3.7-Max88.0
Claude Fable 584.0
Sonnet 584.0
DeepSeek-V4-Pro84.0
Opus 4.580.0
Opus 4.680.0
DeepSeek V4.1 Flash80.0
Gemini 3.5 Flash80.0
Gemini 3.6 Flash80.0
Grok 4.380.0
Grok 4.680.0
MiniMax M380.0
Muse Spark 1.280.0
Opus 4.776.0
Gemini 3 Pro Preview76.0
GLM-5.176.0
Loading Atlas data…