Development and Safety Validation of a High-Performance Japanese Medical LLM

As part of NEDO’s project “Safety Verification and Demonstration for the Social Implementation of a Japanese Medical Domain-Specific LLM,” a collaborative research team comprising ten institutions, including the University of Tokyo, RIKEN, and Kyushu University, developed a high-performance Japanese large language model (LLM) designed to support administrative tasks in healthcare settings. In evaluations based on specialist medical examinations, the model achieved an accuracy of up to 90.8%, demonstrating performance comparable to state-of-the-art commercial LLMs. The project also conducted extensive safety assessments tailored to the characteristics of the Japanese healthcare system.

Jun Seita, Team Director of the Medical Science Deep Learning Team at iTHEMS, also participated in the project and contributed to its research and development.

For more details, please see the NEDO press release.