Which models can actually do CAD?

partforge is an agentic coding harness for parametric CAD. An AI agent works with you to turn a prompt, and whatever guidance follows it, into a small program that generates a 3D model. Along the way it has the documentation, a reference library of published parts, and web search. After each step of a build it gets screenshots and measurements of the thing it has made. That makes this an unusually direct way to measure how the current models handle vision, tool use and reasoning about 3D geometry at once.

This first baseline ran 5 models over 54 tasks, 675 attempts in all. Each attempt is graded on whether it built, on measured geometry, and on a checklist an AI judge answers from the renders. Click any number to open the attempt behind it.

What we found

Some of it surprised us. Cost and time spread much further than quality did: the field runs from $0.16 to $1.02 a task, and from 1m 17s to 3m 59s. Almost any current model can build a washer or a spacer. Ask for a motor spoke, a clamshell cover, or a lid that has to fit something, and the results vary a lot.

The thing we kept running into is that a person's idea of a good design does not always match the AI judges'. Better ergonomics and simpler forms are what set some models apart, and nothing in the score sees that directly.

This is the first of many runs, and we want to see how DeepSeek V4.1 and other open weight models do. On the strength of these results we have made Gemini 3.8 Flash the default model for partforge users. It costs about a sixth of what GPT-6 Astra does, it is quicker than every model that scored near it, and the human ranking came down in its favour more often than not.