Google's Antigravity 2.0 achieved the highest score in a practical test of AI coding tools generating complex 3D models using OpenSCAD, a text-based parametric modeling language. The benchmark required replicating Rome's Pantheon from reference images, emphasizing spatial geometry and architectural precision. Antigravity 2.0 used real Pantheon dimensions, included the iconic inscription, and replicated the interior coffered ceiling pattern, outperforming rivals like Claude Code and Cursor.

ModelRift's human-in-the-loop approach with Gemini Flash 3.0 also excelled, though it required more time than fully autonomous systems.