Baichuan-M3 - Baichuan Intelligence's next-generation medical augmented language model (open source). - AiBoss

What is Baichuan-M3?

Baichuan-M3 is Baichuan Intelligence's new generation of open-source medical augmented language model. The model ranked first in diagnostic capabilities, medical illusion control, and Healthbench and Healthbench Hard benchmarks, surpassing OpenAI's GPT-5.2 and outperforming human doctors in all test categories. Baichuan-M3 is deeply optimized for medical scenarios, integrating massive amounts of medical literature, clinical guidelines, real medical records, and drug knowledge bases. It possesses accurate disease reasoning, medication recommendations, test result interpretation, and patient communication capabilities.

Main functions of Baichuan-M3

Strong medical reasoning abilityIt excels at handling complex medical problems, demonstrating deep medical reasoning and providing accurate diagnostic recommendations.
Low hallucination rateThe rate of medical hallucinations is only 3.5%, the lowest in the world, ensuring the reliability and credibility of the information output.
End-to-end serious consultation capabilitiesThey can proactively inquire about key medical history and risk signals, just like doctors, and collect complete patient information, significantly outperforming the average level of real doctors.
Medical knowledge integrationIt integrates a vast amount of medical literature, clinical guidelines, real medical records, and drug knowledge bases to provide comprehensive medical knowledge support.
Support the medical application "Bai Xiaoying"Doctors can use it to deduce their diagnostic and treatment approaches, and patients and their families can use the application to better understand the medical logic behind diagnosis, treatment, examination, and prognosis.
Open source promotes ecosystem buildingBy adopting an open-source strategy, we aim to promote the widespread application and ecosystem development of medical AI technology, and accelerate its implementation in scenarios such as primary healthcare, assisted diagnosis, and health management.

Technical Principles of Baichuan-M3

Reinforcement learning optimizationBy comprehensively upgrading the reinforcement learning system and taking the consistency of medical facts as the core training objective, the model's performance on complex medical problems is continuously improved, achieving a leap in capabilities that surpasses existing models.
Hallucination Suppression Training ParadigmBy moving hallucination suppression forward to the model training stage and reconstructing the training process, the reliability and consistency of the medical information output by the model can be ensured, achieving the lowest rate of medical hallucinations globally.
Serious consultation paradigm and SCAN principleThe paper proposes a "serious consultation paradigm" that combines Safety Stratification, Clarity Matters, Association & Inquiry, and Normative Protocol to systematically simulate the doctor's consultation thinking process and achieve precise clinical inquiry.
Dynamic evaluation systemWe will construct a SCAN-bench evaluation system, using real clinical experience as the standard, to dynamically and in multiple rounds evaluate the model from medical history collection to accurate diagnosis, ensuring the model's effectiveness in real medical scenarios.
Native model training methodsThe system employs native model training instead of role-playing prompts, and utilizes a novel SPAR algorithm to enable the model to ask key questions within a limited number of dialogue rounds, ensuring the accuracy and completeness of the consultation.
Multimodal fusionBy combining multiple modalities of data, such as text and images, the model's ability to understand and process medical information is improved, better supporting the needs of complex medical scenarios.

Baichuan-M3 project address

GitHub repository: https://github.com/baichuan-inc/Baichuan-M3-235B
Hugging Face Model Libraryhttps://huggingface.co/baichuan-inc/Baichuan-M3-235B

Application scenarios of Baichuan-M3

Assisting doctors in consultationIt helps doctors quickly collect patient medical history, accurately identify key information, and improve the efficiency and accuracy of consultation.
Popularization of medical knowledgeTo provide patients and their families with medical knowledge interpretation, helping them to better understand diagnostic, treatment, examination, and prognostic information.
Primary healthcare supportTo provide auxiliary diagnostic suggestions in primary healthcare institutions, improve the level of primary healthcare services, and alleviate the problem of uneven distribution of medical resources.
Medical research supportIt provides data processing and analysis support for medical researchers, accelerating the medical research process.
Medical Education ToolsAs an auxiliary tool for medical education, it helps medical students and young doctors train in clinical thinking.
Health Management ConsultingIt provides users with daily health management advice, such as disease prevention and healthy lifestyle.