On January 22, Baichuan Intelligence officially launched Baichuan-M3 Plus, a groundbreaking medical large language model that surpasses its predecessor, M3, by setting a new global benchmark for accuracy and reliability in answering medical queries within professional healthcare contexts. By integrating its proprietary six-source evidence-based verification framework with the robust M3 foundational architecture, M3 Plus has achieved an unprecedented reduction in the hallucination rate—a critical metric for model trustworthiness—to just 2.6%. This figure not only outperforms Open Evidence but also represents the lowest hallucination rate ever recorded in the field of medical AI.

A standout feature of M3 Plus is its innovative "evidence anchoring" technology. Unlike conventional models that may generate conclusions without clear traceability, M3 Plus goes a step further by not only citing the sources of its medical insights but also ensuring a precise, contextual alignment between the AI-generated recommendations and the specific evidence passages within the original peer-reviewed papers. This approach significantly enhances the model's credibility and usability in clinical decision-making processes.