At a critical stage where medical AI is moving towards reliable clinical applications, Baichuan Intelligence officially launched Baichuan-M3 Plus - a large model specifically designed for high-reliability medical scenarios. This model innovatively integrates the "Six-Source Evidence-Based Paradigm," which has been validated in M2 Plus, deeply into the entire training and reasoning process. This ensures that every diagnostic suggestion, medication reminder, or health guidance comes from authoritative medical evidence, reducing the factual hallucination rate to just 2.6%, a reduction of over 30% compared to the current industry-leading GPT-5.2, setting a new record for trustworthiness in medical large models.
The "Six-Source Evidence-Based Paradigm" is Baichuan Intelligence's core methodology in the field of medical AI. It refers to the model's requirement to comprehensively refer to six types of authoritative information sources when generating responses: including international treatment guidelines (such as WHO, NCCN), national pharmacopoeias, core journal literature, clinical trial databases, drug instruction manuals, and real-world research data. By injecting structured evidence-based knowledge during the training phase and dynamically retrieving and verifying during reasoning, M3 Plus effectively avoids the common "confidently making up" risks seen in general large models.
Testing results show that in high-risk scenarios such as complex case analysis, drug interaction judgment, and chronic disease management advice, M3 Plus not only significantly improves answer accuracy but also actively marks the information sources, supporting doctors to trace back with one click. This capability upgrades it from an "information provider" to a "auditable intelligent collaborator," greatly enhancing clinical confidence.
Baichuan Intelligence stated that M3 Plus will be prioritized for release to hospitals, online medical platforms, and pharmaceutical R&D institutions. In the future, it is expected to be integrated into electronic medical record systems, remote consultation platforms, and clinical decision support tools. In the medical field, where the tolerance for errors is extremely low, "saying less but being accurate" has become the core criterion for AI implementation. The release of Baichuan-M3 Plus marks a crucial transition for domestic large models in specialized vertical fields, moving from "usable" to "trustable."
