MUSE: Benchmarking Manufacturable, Functional, and Assemblable Text-to-CAD Generation

Researchers introduced MUSE, a new Text-to-CAD benchmark for complex editable CAD assemblies.

It evaluates outputs with a three-stage pipeline: code checking, geometry checking, and design-intent alignment.

The benchmark uses structured design specifications and a rubric-based VLM judge, validated against human annotations.

Results show current LLMs still struggle to generate engineering-ready CAD that is functional, manufacturable, and assemblable.