[Context and Motivation] User stories are informal, natural language descriptions that outline features from a user’s perspective, guiding collaboration and iterative development in agile projects. [Question/problem] Engineers typically interpret these informal descriptions manually to create domain models—a process that is time-consuming and reliant on their expertise. Recent advancements in Large Language Models (LLMs) show promise in automating such tasks through improved language understanding, generation, and reasoning abilities. [Principal ideas] This paper investigates the potential of open-source LLMs to automatically generate domain models from NL requirements, employing instruction tuning on a substantial dataset of user stories and their corresponding domain models. We compare instruction-tuned and pre-trained models to determine the specific impact of instruction tuning on model generation quality. [Contributions] To our knowledge, this is the first study applying instruction tuning on LLMs specifically for domain model generation. Through both qualitative and quantitative analyses, assessing factors such as completeness and correctness, we evaluate these models’ effectiveness across user groups, from students to expert analysts. This evaluation helps identify the most suitable user groups that could benefit from this technology in both educational and professional settings.