AI BriefingAntigravity 2.0 Tops OpenSCAD Architectural 3D LLM Benchmark
Key point
Google Antigravity 2.0 and Gemini 3.5 Flash High achieved the highest quality scores in an OpenSCAD-based architectural 3D modeling benchmark.
Details
The OpenSCAD Pantheon Benchmark, conducted by ModelRift, evaluates the ability of AI coding tools to implement architectural structures as parametric CAD code using only reference images and short prompts.
Key results are as follows:
- Google Antigravity 2.0 / Gemini 3.5 Flash High: Ranked first with a quality score of 4.5/5. It reproduced the Pantheon's actual dimensions and inscriptions, and was the only autonomous agent to also implement the interior coffered ceiling pattern.
- Claude Code 2.1 / Sonnet 4.6: Took the longest implementation time, but showed the most balanced proportions and impression among the autonomous execution models, scoring 3.4/5 in quality.
- Codex 5.5 High: Had high detail density, but lost points due to a mismatch between the PNG preview and the final STL file.
- Cursor: Was the fastest in implementation speed (5/5), but had the lowest quality at 1.4/5.
This test confirmed that the performance bottleneck for AI models lies not simply in tool access, but in complex geometric judgment and the ability to validate the final mesh.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.