Model Rift pits CadQuery against OpenSCAD in a head-to-head test of AI-generated functional parts. Six autonomous agents, powered by Claude Opus 5, completed three design tasks using CadQuery 2.8.0 and OpenSCAD 2026.06.12 on M-series Macs. All six resulting STL files passed rigorous geometric validation, confirming that capability is not the differentiating factor.

The study reveals that the tools diverge in their failure modes. In the most complex task, an M24x2 threaded hose-barb adapter, both tools succeeded, but their approaches differed significantly. OpenSCAD leverages its established skill set and CSG status line, while CadQuery requires an offscreen renderer and B-rep validity metrics.

A key asymmetry exists: CadQuery files include an API cheatsheet, benefiting syntax, though this offers no geometric advantage. Summed metrics show OpenSCAD averaged 77k tokens and 461 seconds per task, while CadQuery averaged 119k tokens and 909 seconds. The results suggest that while both tools are capable of producing printable parts, their underlying architectures and error handling strategies produce distinct developer experiences when tasked with complex functional geometry.