This paper presents a comprehensive survey of how to apply pre-trained, large-scale language models (foundation models) to the Internet of Things (IoT). To address the challenges of existing machine learning approaches, which often suffer from data insufficiency and overfitting to specific tasks, we focus on the advantages of foundation models, which can be generalized to a wide range of tasks. Unlike previous studies that focus on specific IoT tasks, this paper systematically categorizes and analyzes existing research around four common performance objectives: efficiency, situational awareness, safety, security, and privacy. For each objective, we review representative studies and summarize commonly used techniques and evaluation metrics, enabling meaningful comparisons across IoT domains and providing practical insights for selecting and designing foundation model-based solutions for new IoT tasks. Finally, we suggest future research directions and offer guidelines for practitioners and researchers to advance the use of foundation models in IoT applications.
