Quark Releases Health Large Model Technical Report, Disclosing Technical Details of AI 'Chief Physician'
-
Recently, Quark officially released the health large model technical report 'QuarkMed Technical Report,' for the first time disclosing the technical implementation details of its 'chief physician-level' capabilities.
Previously, the Quark Health Large Model successfully passed the written examination for chief physicians in 12 core disciplines in China, becoming the first domestic large model to achieve this milestone. Compared to general models, the Quark Health Large Model exhibits a performance curve where 'the higher the difficulty, the more pronounced the lead,' particularly achieving breakthroughs in complex medical reasoning tasks. This technical report systematically reveals the key pathways and technical highlights behind this breakthrough.
Addressing the challenge of medical models requiring high-quality, highly specialized training data, the Quark Health Large Model utilizes three core types of medical data at different stages of model training: medical literature, medical knowledge, and medical records, totaling approximately 1 trillion tokens. This professional data effectively compensates for the shortcomings of pre-training corpora, enhancing the model's accuracy and reasoning capabilities.
To improve the model's correctness, safety, and complex reasoning abilities, the Quark Health Large Model incorporates two reinforcement learning (RL) phases. The first phase employs large-scale medical reinforcement learning to enhance the model's reasoning capabilities in complex scenarios. The second phase designs a reward model to evaluate the quality of model outputs from three perspectives—honesty, usefulness, and content compliance—adjusting the model's behavior to align with human preferences and values.
The technical report also presents multiple performance test results. In evaluations on authoritative international datasets such as MedQA, the Quark Health Large Model demonstrated superior performance compared to models of similar size like o3-mini. In the Chinese Physician Qualification Examination (CPQExam) written test, the higher the difficulty, the more pronounced the Quark Health Large Model's lead became.
The report reveals that Quark plans to fully disclose the physician examination test set to promote AI research in the medical field.
Access and download links:
https://arxiv.org/pdf/2508.11894
https://github.com/Quark-Medical/QuarkMed/blob/main/report/QuarkMed_Technical_Report.pdf