Samsung Medical Center's Medical AI Achieves World-Class Performance on Korean and U.S. Medical Licensing Exams

KMLE Score 97.7, USMLE 94.8: Achieving SOTA Performance
Advancing Clinical Application Capabilities

The medical AI solution "GeomMed2," developed by Samsung Medical Center, achieved state-of-the-art (SOTA) performance in evaluations based on the Korean and U.S. national medical licensing examinations.

GeomMed2 demonstrated its clinical reasoning ability by correctly selecting the appropriate treatment drug, lorazepam, in actual medical licensing test scenarios, such as emergency care for pediatric patients experiencing high fever and recurrent convulsions.

Professors Chun Chul Cho and Myunghee Son at Samsung Medical Center. Samsung Medical Center

On July 29, Samsung Medical Center announced that a research team led by Professors Chun Chul Cha and Myunghee Son of the Department of Emergency Medicine, in collaboration with the AI specialist firm ZeroOneAI, developed GeomMed2 through the Advanced GPU Utilization Support Project under the Ministry of Science and ICT.

According to the hospital, GeomMed2 recorded a score of 97.7 on the Korean Medical Licensing Examination (KMLE) and 94.8 on the U.S. Medical Licensing Exam (MedQA-USMLE), achieving the highest scores among competing models in benchmark performance tests.

The research team released model information, performance on each benchmark, evaluation logs, and reproducibility verification codes on the AI model-sharing platform Hugging Face, making them accessible for public use.

Samsung Medical Center plans to further advance model performance using clinical data and add multimodal capabilities to process integrated information such as voice and images. The hospital will also strengthen safety and control features required in clinical settings through harness engineering, and pursue model lightweighting to broaden real-world clinical applicability.

Professor Cha stated, "This achievement was made possible by the combination of government GPU support, the hospital's clinical experience and data, and the technological expertise of the AI company. We aim to develop it beyond merely solving exam questions, turning it into a clinically specialized AI solution that can be used in actual medical practice."

This content was produced with the assistance of AI translation services.