Core Viewpoint - The article highlights the significant achievements of the Run Medical Model, developed based on the Pangu large model, in the field of AI and healthcare, establishing a new benchmark for AI+medical applications [1][6][10]. Summary by Sections MedBench Evaluation - The Run Medical Model topped the MedBench evaluation platform with scores of 96.4 and 93.2 in the professional and self-assessment categories, respectively, showcasing its leading position in the medical AI sector [3][4][8]. - In the self-assessment category, the model excelled in four dimensions: medical knowledge Q&A (91.2), medical language generation (85.1), medical language understanding (123.1), and medical safety and ethics (106.6) [4][8]. - The model also performed well in the professional evaluation, achieving scores of 87.7, 84.8, 122.4, and 98.7 in the same respective categories [4][8]. Technical Innovations - The Pangu model team utilized a vast dataset, including hundreds of billions of high-quality Chinese and English medical literature, guidelines, and millions of health records, to enhance the Run Medical Model's medical knowledge and expression capabilities [8][9]. - A multi-agent medical data synthesis workflow was introduced to improve the model's ability to capture complex patterns and relationships in medical data, enhancing its Q&A, language generation, and reasoning capabilities [9][10]. Industry Impact - The advancements in the Run Medical Model are seen as a significant milestone in the development of medical AI, reflecting the team's deep technical accumulation and innovative strength in the medical vertical [8][10]. - The Pangu model serves as a powerful foundation for medical AI, driving the industry towards a new era of intelligent, precise, and personalized healthcare [10].

Group 1 - MedBench released a new evaluation ranking on May 27, where DeepBlue-MR-v1 from DeepBlue Technology ranked first in complex medical reasoning and achieved a high score of 94.2 in multiple comprehensive evaluations [1][2] - The evaluation platform MedBench is recognized as the leading authority in Chinese medical large model assessments, established by the Shanghai Artificial Intelligence Laboratory and the Shanghai Digital Medicine Innovation Center, and has evaluated over 387 models globally [1] - Other models evaluated include Tencent's YouTu Tianyan Medical Model, Huawei's Pangu-based Runyi Medical Model, and Yunzhisheng's UniGPT-Med-U1 [1] Group 2 - DeepBlue-MR-v1 is a self-developed medical reasoning model by DeepBlue Technology, excelling in clinical medical inquiries, assisting in medical diagnoses, and formulating treatment plans [2] - The model utilizes a vast dataset including medical textbooks, treatment guidelines, expert papers, and case histories, employing a self-developed training system to align human medical reasoning capabilities with a Transformer-based dense language model [2] - DeepBlue-MR-v1 has maintained its leading position in complex medical reasoning and has also topped the MedBench rankings in five dimensions: medical language understanding, medical language generation, medical knowledge Q&A, complex medical reasoning, and medical safety and ethics [2] Group 3 - Based on the DeepBlue-MR-v1 model, DeepBlue Technology has developed a product matrix that includes "AI Inquiry Assistant," "Remote Video Consultation," "Auxiliary Diagnosis System," and "Medical Expert Knowledge Base" [3] - The company has established partnerships with several medical institutions, including Wuhan Central Hospital and Wuhan Union Hospital, to promote the deep application of AI technology in inquiry, diagnosis, and specialized services [3]