Baichuan Intelligence on Thursday unveiled its enhanced medical large language model, Baichuan-M3 Plus, claiming improvements in accuracy and reliability for serious medical question-and-answer scenarios over its recently released M3 model.

The company said that by combining its six-source evidence technology with the M3 base model, M3 Plus reduces hallucination rates to 2.6%, lower than the benchmark set by Open Evidence. The model introduces a novel “evidence anchoring” feature, which not only provides citation sources but also links each AI-generated medical conclusion to the corresponding evidence paragraph in the original research paper, making medical judgments verifiable, accountable, and suitable for teaching purposes.

In addition, Baichuan announced the launch of its “Hai Na Baichuan” (All Rivers) initiative, offering the evidence-enhanced medical model via API free of charge to Chinese healthcare institutions.