CADWorld Benchmark for Long-Horizon Computer-Aided Design
September 13, 2026
CADWorld is a benchmark for computer-use agents operating in FreeCAD, covering 200 tasks across 11 categories like sketching and assembly. It evaluates an agent's ability to produce valid, structured engineering artifacts through screenshot and interface interaction.
HOW THIS AFFECTS YOU
●
builderYou can test your computer-use agents against professional-grade engineering workflows that require geometric consistency.
●
researcherYou can use this benchmark to evaluate how well agents handle long-horizon, structured artifact generation.