Seoul National University Hospital has developed something that can be Korea’s first large language model.
The project to develop medical LLM began in March last year. The SNUH Research Team started with non -identification and anonymity to collect 38 million clinical texts, including hospital hospitalization, outpatient patients, surgery, prescriptions and nursing records, to create a default copper for model learning.
Then, in early 2025, we developed a knowledge foundation for each department and later integrated into LLM through search sedimentation (RAG). This knowledge base consists of medical terminology and abbreviations, as well as local medical law, paper green and treatment guidelines.
LLM has been tested for questions in the last three years of the Korea Medical License Test and has recorded 86.2%, surpassing 79.7%of the average takers. SNUH also said that this model showed high translation performance that handled 50,000 words at the same time.
The SNUH Research Team will check the performance and safety of LLM for one year before applying to support clinical work and research. In addition, we plan to expand applications in various medical fields and further improve the medical data processing function of the model.
Why is it important
According to SNUH, the existing medical LLM developed by a technology giant including Google MED-PALM 2 and Microsoft’s LLava-Med are limited to understanding Korean medical text, law and treatment guidelines. These models are optimized in the Western context, so it has been found to be incomprehensible to Korean medical professionals. As a result, hospital researchers began to develop Korea Medical LLM to meet the needs of local clinical trials.
Greater trend
At the end of last year, startups, supported by large corporations SK Group, insisted that they introduce the first LLM -based search platform for Korea’s medical services. PHYNX LAB CHEIRON supports pharmacists and pharmaceutical industry researchers to search for relevant information by selecting the user’s intention through natural language processing. You can also review your own information and create high quality answers quickly.
Korea University ANAM Hospital It is also developing its own LLM, which is expected to be controlled this year.
Asan Medical Center In addition, LLM is used to create voice -based clinical scribes and is currently used in 16 departments.
Meanwhile, SNUH is currently conducting other AI projects and free clinical trials in administrative work. This includes multimodal AI that can automatically generate patient summaries, AI to select the latest papers for researchers to strengthen insurance claims.
Record
KIM Young-TAE, the president and CEO of SNUH, said, “The development of this Korean medical LLM maximizes the work efficiency of medical staff and provides faster and more accurate medical services to the patients.
