Abstract: With the advancement of intelligent coal mine construction, artificial intelligence technologies have been deeply applied in areas such as production management, equipment maintenance, safety inspection, and geological exploration. However, existing AI models often suffer from limited parameter size, low accuracy, and single-mode operation, which hinder their ability to meet the complex task requirements of coal mines and restrict industry progress. The emergence of large-scale models has brought significant improvements in language understanding and reasoning capabilities, offering great potential for accelerating intelligent coal mine development and promoting safe and efficient mining practices. Large models mainly include language-based, vision-based, multimodal, and other types of models. These models are typically based on the transformer architecture and leverage techniques such as data augmentation, chain-of-thought prompting, RLHF and LoRA fine-tuning to enhance performance and flexibility. This paper reviews the current research and application scenarios of large models both domestically and internationally, despites their promising capabilities, large models still face significant challenges in mining environments, which pose higher requirements on model training and generalization. Future research should focus on improving the spatiotemporal prediction capabilities of large models in mining scenarios, exploring mixture of experts architectures, and advancing embodied intelligence technologies for mining equipment to accelerate practical deployment. These efforts will provide a solid theoretical and technical foundation for intelligent coal mine systems driven by large models, promoting industrial transformation and technological innovation in the mining sector.