LLMs are surprisingly good at generating CAD models

Recent research demonstrates the surprising ability of Large Language Models (LLMs) to generate CAD models for simple 3D mechanical parts, with performance rapidly improving. An engineer combined an LLM with the open-source programmatic CAD tool OpenSCAD, successfully generating models like an iPhone case using natural language prompts. A subsequent evaluation framework, CadEval, tested various LLMs' CAD generation capabilities, revealing that reasoning models significantly outperform their non-reasoning counterparts. Startups are also entering the text-to-CAD space, but their performance currently lags behind the LLM-OpenSCAD approach. Future advancements in LLMs and related technologies promise widespread adoption of text-to-CAD in mechanical engineering, ultimately automating and intelligently enhancing CAD design.