Recently, the Baichuan large model officially released its latest medical large model - Baichuan-M2Plus, and upgraded the accompanying application "Bai Xiao Ying" at the same time, opening up API interfaces. This release marks another important progress since Baichuan-M2 was open-sourced in August. Through evaluations, M2Plus has shown excellent performance in reducing medical hallucination rates, significantly lower than general large models, especially compared to existing medical products such as DeepSeek, with a hallucination rate reduced by about three times, even surpassing the popular medical application OpenEvidence in the United States.
M2Plus, with its innovative six-source evidence-based reasoning (EAR) paradigm, has become the "ChatGPT for doctors". The model addresses application challenges in serious medical scenarios by integrating original research, evidence reviews, guidelines, practical knowledge, public health education, and regulatory real-world information, building a complete medical knowledge system, ensuring the credibility and scientific nature of medical decisions.
In terms of evidence-based retrieval, M2Plus uses the PICO framework to convert medical queries into structured questions, ensuring that the retrieved information is accurate and reliable. The unique design of this model also allows doctors to obtain high-level, credible medical evidence when facing complex medical problems, greatly improving the efficiency of using medical information.
More importantly, the "evidence-enhanced training" mechanism in M2Plus when answering medical questions ensures that the model's answers are not just based on retrieved evidence but effectively avoid generating random information. By reinforcing the use of authoritative sources and evaluating the quality of evidence, M2Plus demonstrates a level of credibility comparable to that of experienced clinical experts in practical applications.
In multiple tests in medical scenarios, M2Plus has received high praise from clinical doctors, especially in analyzing medical histories, diagnostic thinking, and treatment plans, with its answers being widely recognized for accuracy and professionalism. More impressively, M2Plus achieved an impressive score of 97 in the United States Medical Licensing Examination (USMLE), demonstrating its superior ability in applying medical knowledge.
The release of the Baichuan large model not only injects new momentum into the digital transformation of China's medical industry, but also represents further breakthroughs of China in the global large model field. With the launch of M2Plus, it is expected to have a profound impact on improving the scientific and effective nature of clinical decision-making.
