https://www.mdu.se/

Large Language Models such as ChatGPT have recently received significant traction in software development. Many papers are dedicated to investigating and understanding the capabilities and shortcomings of Large Language Models in code generation. However, the same cannot be said for its use within the domain of software modeling. The exploration of leveraging Large Language Models for software modeling tasks is yet to be analyzed further. In this paper, we explore the abilities of ChatGPT in conducting model transformations, focusing on translating Unified Modeling Language class diagrams encoded in XML Metadata Interchange format to object-oriented code. In conducting the experiment, we developed a pipeline to evaluate the performance of ChatGPT to handle model transformations based on different complexity levels. Our results indicate that while Large Language Models like ChatGPT currently prove to be effective in handling very simple to intermediate model transformations, its performance declines with increased complexity of the models, resulting in errors like missing classes and incorrect types, among others.