Answers are grounded in this paper and cite their sources.
The use of generative artificial intelligence (AI) is rapidly increasing across various sectors, making tasks more accessible, efficient, and user-friendly. One emerging application is computer-aided design (CAD), where AI has the potential to enable users with little or no-prior CAD experience to create three-dimensional (3D) models through natural-language. Despite growing interest in AI-assisted design, the capability of publicly accessible generative AI systems to reliably generate CAD models across varying levels of design complexity remains insufficiently studied. This work benchmarks the performance of Microsoft Copilot for CAD model generation and investigates the influence of prompt engineering on model quality. A benchmark dataset comprising 60 design tasks was developed and categorized into easy, intermediate, and complex levels. The benchmark included primitive geometries, polygonal and polyhedral models, mechanical components, hollow structures, assemblies, lattice designs, and advanced engineering systems. For each prompt, the generated STL file was evaluated based on geometric accuracy, completeness of requested features, and overall structural validity using commercial software. The results revealed a strong dependence of generation success on design complexity, with easy-level tasks exhibiting high first-attempt success rates, while intermediate and complex tasks frequently required prompt refinement. Prompt length alone was found to have minimal influence on performance; instead, successful CAD generation depended primarily on clearly specified dimensions, geometric features, and functional relationships. Overall, the findings demonstrate that generative AI can lower the barrier to CAD modeling for non-expert users while improving productivity for experienced designers, thereby offering significant potential for engineering design, rapid prototyping, and additive manufacturing applications.
No takes yet. Share an insight, caveat, or question.
Lara et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: