LLMs for Model-driven Engineering: A Systematic Literature Review
The paper was published online in ACM Transactions on Software Engineering and Methodology on October 21, 2026. This systematic literature review analyzes 235 primary studies covering 25 types of model-driven engineering tasks, with model generation as the dominant focus. It examines LLM techniques, benchmarking and evaluation practices, and the availability of tools and datasets. The findings show that UML, DSLs, and BPMN are the most frequently targeted formalisms. GPT-family models, zero-shot prompting, API-based access, and single-LLM architectures prevail, with fine-tuning and retrieval-augmented generation as common enhancements. Evaluations focus mainly on correctness, with limited realistic benchmarking and insufficiently comprehensive replication materials.