Antigravity 2.0 Tops OpenSCAD Architectural 3D LLM Benchmark
Original: Antigravity 2.0 Tops the OpenSCAD Architectural 3D LLM Benchmark
Why This Matters
Demonstrates AI progress in generating complex 3D architectural models from text
ModelRift conducted a benchmark testing six AI coding tools on building the Pantheon in OpenSCAD. Google Antigravity 2.0/Gemini 3.5 Flash High performed best, followed by ModelRift/Gemini Flash 3.0 and Codex 5.5 High.
ModelRift tested six AI systems' ability to generate OpenSCAD code for architectural models using the Pantheon as a benchmark. The test evaluated Google Antigravity 2.0/Gemini 3.5 Flash High, ModelRift/Gemini Flash 3.0, Codex 5.5 High, Claude Sonnet, Cursor Composer, and Claude Opus. Each system received the same visual prompt to build the Pantheon including rotunda, dome, portico, columns, and pediment from reference images. OpenSCAD was chosen as the target because its text-based parametric approach aligns well with how language models process structure through Boolean operations and transformations. The Pantheon benchmark tests spatial geometry understanding beyond basic OpenSCAD syntax, requiring systems to handle radial symmetry, extrusions, and architectural relationships between components.