Performance of GPT-5, DeepSeek, and Claude in dental MCQs for medically compromised patients.

backgroundArtificial intelligence (AI) has shown remarkable potential in medical education and clinical decision support, yet its role in dentistry—particularly in the management of medically compromised patients—remains largely unexplored. No previous study has systematically benchmarked the performance of advanced large language models (LLMs) in this high-risk domain.

objectiveThis study provides the first comparative evaluation of three LLMs—GPT-5, DeepSeek, and Claude—on multiple-choice questions (MCQs) specifically designed for dental management of medically compromised patients.

methodsA total of 72 validated MCQs were constructed from the 10th edition of Little & Falace’s Dental Management of the Medically Compromised Patient, covering 18 systemic conditions relevant to dental practice. Each model was independently assessed under identical conditions. Accuracy, agreement with the gold standard answers from the textbook, and error patterns were analyzed.

resultsGPT-5 achieved the highest accuracy (90.28%), followed by Claude (88.89%) and DeepSeek (87.50%). Performance varied across systemic conditions, with all models demonstrating lower accuracy in complex scenarios such as infective endocarditis and bleeding disorders. Qualitative analysis revealed differences in reasoning depth, error types, and consistency of responses.

conclusionsThis is the first study to benchmark multiple frontier LLMs in dentistry, focusing on medically compromised patients—a group where safe and accurate decision-making is essential. The findings highlight both the promise and limitations of AI in dental education and clinical guidance. By systematically identifying strengths and weaknesses, this work provides an evidence base for integrating LLMs into dental curricula and decision-support systems, while underscoring the need for human oversight in complex medical cases. CLINICAL SIGNIFICANCE: Generative AI models, including GPT-5, DeepSeek, and Claude, demonstrated high accuracy in case-based dental decision-making for medically compromised patients. Their integration could enhance dental education and clinical support. However, variability in performance underscores the need for cautious use and further validation before applying AI tools in complex patient care.