The GigaChat neural network is advancing rapidly and expanding into new knowledge domains, according to Sergei Zhdanov, who leads the Sberbank Health Industry Center. The model has demonstrated its growing capability by completing a General Medicine program at a medical school level, a milestone that positions it among the first AI systems worldwide to earn such a credential. This achievement marks a notable step in merging artificial intelligence with practical medical training requirements.
Sberbank announced that the AI performed at the level expected of a medical school graduate, receiving a final grade of 4 on an exam ticket. A commission comprising professors of therapy, surgery, and obstetrics-gynecology from the National Center for Medical Research Institute of Medical Education, VA Almazova, evaluated the work. The examination simulated real-world clinical decision making through a standard oral ticket consisting of three situational tasks across therapy, surgery, and obstetrics-gynecology, complemented by 3–5 follow-up questions for each area. In addition, the model successfully tackled a 100-question written test, achieving 82% with a passing threshold of 70%.
Zhdanov expressed appreciation for the role of the Almazov Center staff, who supervised the training and validation process. He noted that passing the doctor’s examination represents a completed first stage and foresees future work to further integrate GigaChat into healthcare workflows.
Looking ahead, the developers envision the neural network serving as a foundation for tools that assist both physicians and patients, delivering personalized health information and guidance. The aim is to use large language models and their successors as key technologies in advancing human-centered care, improving access to medical knowledge and support for individuals seeking health information.
Almazov Center’s general director and Evgeniy Shlyakhto, president of the Russian Society of Cardiology, described the GigaChat medical training initiative as a major institutional challenge that mobilized hundreds of educators and researchers. The project gathered active participation from citizens and students, and while current results are encouraging, ongoing training and refinement are planned to expand the model’s capabilities and reliability in clinical contexts.
Shlyakhto also indicated that several GigaChat–based applications for medical institutions, patients, and clinicians are already in the pipeline with development set to begin within the year, in collaboration with Sber. This signals a broader strategy to embed AI into the everyday operations of healthcare while maintaining rigorous validation and oversight.