Evaluating the Performance of DeepSeek 3, Claude Sonnet 4, and Gemini 2.5 in the Chilean Medical Licensing Examination: Observational Study.

Introduction: Artificial intelligences and their continuous improvement have revolutionized medical education, but their performance in specific evaluative contexts still requires further exploration. Methods: This study qualitatively evaluated and compared the performance of three state-of-the-art...

Descripción completa

Detalles Bibliográficos
Autores: Jerez Yañez, Oscar, Edgardo, Vicente Alberto, Silva Arroyo, Jesús, Vera Cartes, Marcos Jeremías Giovanny, Herrera Alcaíno, Alvaro Andrés, Lancellotti Guajardo, Anaís Aracelly
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Universidad de Murcia
Repositorio:DIGITUM. Depósito Digital Institucional de la Universidad de Murcia
OAI Identifier:oai:digitum.um.es:10201/201202
Acceso en línea:https://doi.org/10.6018/edumed.679731
http://hdl.handle.net/10201/201202
Access Level:acceso abierto
Palabra clave:Medical education
EUNACOM
Clinical reasoning
Language models
Medical assessment
Inteligencia artificial
Educación médica
Razonamiento clínico
Modelos de lenguaje
Evaluación médica
Artificial intelligence
No relacionado con ningún objetivo de desarrollo sostenible
Descripción
Sumario:Introduction: Artificial intelligences and their continuous improvement have revolutionized medical education, but their performance in specific evaluative contexts still requires further exploration. Methods: This study qualitatively evaluated and compared the performance of three state-of-the-art language models — Claude Sonnet 4, Gemini 2.5, and DeepSeek 3 — in simulations of the National Medical Knowledge Examination (EUNACOM) in Chile. Three mock exams with 180 questions each were used, covering various medical areas and question types, including those based on clinical cases. Results: The results show that all AI models consistently passed the exams, with Claude Sonnet 4 achieving the highest overall performance (89% accuracy) and the greatest consistency across attempts. Clinical case-based questions were answered more accurately than theoretical knowledge questions, highlighting the models' strength in contextual clinical reasoning. Claude excelled in Internal Medicine and Psychiatry, DeepSeek in Surgery, and Gemini demonstrated balanced performance. However, specific gaps were identified in areas such as Public Health and clinical follow-up, suggesting the need for model-specific adjustments. Conclusion: The findings support the educational potential of these tools but also emphasize the importance of their ethical, supervised, and complementary use alongside traditional medical training. This study contributes to understanding the emerging role of artificial intelligence in professional assessments, as well as its limitations and opportunities within the Chilean medical context.