<p>Large language models (LLMs) have emerged as potential tools to support clinical decision-making, patient education, and research. However, their reliability in specialty-specific contexts such as otorhinolaryngology remains underexplored. This study aimed to evaluate the performance of three publicly available LLMs—ChatGPT-4, Gemini-Flash 2.5, and Grok-3—in responding to otorhinolaryngology clinical vignettes across multiple domains. Sixty clinical vignettes were developed covering four domains: Diagnosis, Management, Management of Complications, and Prevention/Control, equally distributed across otology, rhinology, and laryngology. Responses generated by each model were evaluated independently by subject-matter experts using the modified DISCERN-AI (mDISCERN) tool on a five-point scale. Descriptive statistics were applied to compare model performance across domains and subspecialties. Gemini achieved the highest overall mean mDISCERN score (4.77), followed by ChatGPT (4.57) and Grok (4.53). All models performed strongly in the Diagnosis and Prevention domains, with Gemini and Grok scoring 4.80 in both. Gemini excelled in Management (4.90) and achieved a perfect 5.0 in Management of Complications, whereas Grok underperformed in these domains, with several low-scoring responses. Subspecialty analysis revealed that Gemini performed best in otology (4.90) and laryngology (4.80), while ChatGPT had its highest scores in rhinology (4.65). Gemini demonstrated superior performance and consistency across domains, followed by ChatGPT. Grok’s lower scores, particularly in management-related tasks, underscore the need for caution when using LLMs for clinical decision support. Ongoing validation and transparency measures are essential prior to routine clinical integration.</p>