Objectives: Generative AI has emerged as a mature technology poised to enhance efficiency in various fields traditionally considered exclusive to human expertise. Computer-Aided Design (CAD) is one such domain that could benefit from this technological evolution. This study explores the potential of Large Language Models (LLMs) in generating code for creating 3D models through text prompts, specifically in CATIA using its scripting module. Materials and Methods: A series of modeling challenges were defined with increasing levels of complexity. LLMs were prompted to generate scripting code capable of constructing corresponding 3D models in CATIA. The generated scripts were tested, and their outputs were analyzed based on accuracy, completeness, and the need for manual intervention. Results: The experiments revealed that while LLMs can generate functional code for 3D modeling tasks, manual review and refinement are often required to ensure correctness and usability. The study also highlighted limitations in model comprehension of geometric constraints and parametric design principles. Conclusions: This research highlights the potential of integrating reinforcement learning techniques into CAD scripting. By controlling 3D modeling through scripts it is possible to design automated processes that adjust parametric models dynamically. Future work could explore how AI-driven optimization strategies can enhance geometric modeling and automate complex design decisions.