Reproduce “Parametric CAD Bench v3: Evaluating Agentic Parametric CAD Authoring from Text and Engineering Drawings” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Mei Chen, Yu Xia, Haipeng Mao

The benchmark evaluates agents that create native FreeCAD models from text, edit their own models, or reconstruct parts from engineering drawings. It scores geometry similarity and dimensions and relationships measurable in the CAD document, combining them into continuous task rewards. In a 100-task snapshot, Claude Code with Opus 5.5 reaches 61.03% overall and 84.66% on drawings; these are mean rewards, not perfect success rates.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.