GPT 6 or Claude Opus 5.5 for SolidWorks CAD automation via MCP?
A user automating SolidWorks CAD work through MCP asks which model to commit to before training seriously on their own data. The practical answer from the thread: run a blind shootout on 5-10 of your own real drawings with identical tools and instructions. Model choice reportedly matters less than MCP server quality and chunking the task to one feature per agent step.