In the field of ship engineering, a 3D sample model is a simplified 3D digital model representing the basic geometric shape and general structure of an actual object. It is often used to verify and optimize the structure and performance of ships. Traditional 3D sample models are represented as a series of computer-aided design (CAD) operations, which are complex and rely on specialized modeling knowledge. Although large language models (LLM) perform well in various fields, their application in 3D modeling may lead to inaccuracies and robustness issues due to the phenomenon of 'hallucinations.' Therefore, a large model-driven enhanced 3D sample modeling method for ships is proposed, using a parametric modeling instruction set to represent the 3D sample model. First, based on a retrieval-enhanced generation method, existing 3D sample model code instructions are searched and invoked, using large model tools to drive the parameters defined in the model, reducing hallucinations in generating 3D sample models. Second, an external ship professional knowledge base is introduced to supplement the generated 3D model with information, adding features, process requirements, etc., making the 3D sample model more useful in downstream applications. The research results show that the large model-driven enhanced 3D sample modeling can effectively handle different modeling requirements, quickly generating 3D sample models with up to 15 features that meet the requirements, significantly improving modeling efficiency.