check this out: https://modelrift.com/blog/openscad-llm-benchmark LLMs are getting pretty good at OpenSCAD, especially if they work in agentic mode and can inspect intermediate results.