Abstract: Objective To investigate the current status of medical students' use of large models in the digitaltransformation of medical education, and to provide data support for optimizing teaching resources. Methods Aquestionnaire survey was conducted using a 5-point Likert scale. A total of 160 medical undergraduates in grades 3 to 5from a medical college were surveyed, covering 7 dimensions including usage coverage, effectiveness in learning/research/clinical scenarios, and training needs. Results The penetration rate of large models exceeds 95%. Among medicalstudents, 32% use them more than 10 times per week, with domestic models such as DeepSeek taking the lead in usage rate.In learning scenarios, information retrieval (88. 75% high effectiveness) and text translation (86. 25% high effectiveness)perform prominently, while knowledge point mastery (43. 75% low effectiveness) needs improvement. In scientific researchscenarios, literature analysis shows significant efficiency (85. 63% high effectiveness), but the assistance in experimentaldesign is limited. Medical record writing stands out with remarkable results, taking the lead with a 40. 62% dataperformance and showing an obvious unimodal skewed distribution "characteristic". "Case discussion" has the highestproportion of negative evaluations, with the combined proportion of "no effect" and "slightly effective" reaching 40%. For"error correction", the proportion of negative evaluations is 62. 50%. 75% of students demand training, focusing on practicaloperations and scenario applications. Conclusion Large models have become important tools for medical students' learningand research, but they have limitations in knowledge internalization and clinical thinking cultivation. It is necessary tooptimize functions and carry out targeted training based on institutional needs.