Baichuan-M3: Modeling Clinical Inquiry for Reliable Medical Decision-Making

TL;DR

Baichuan-M3 is a specialized medical large language model designed for active clinical decision support, featuring proactive information gathering, long-term reasoning, and hallucination suppression, achieving state-of-the-art results in medical inquiry tasks.

Contribution

It introduces Baichuan-M3, a novel medical LLM that models physician workflows for improved clinical decision-making and outperforms existing models on new medical benchmarks.

Findings

Achieves state-of-the-art results on HealthBench, HealthBench-Hallu, and ScanBench.

Significantly outperforms GPT-5.2 in clinical inquiry and safety.

Models are publicly available for research and development.

Abstract

We introduce Baichuan-M3, a medical-enhanced large language model engineered to shift the paradigm from passive question-answering to active, clinical-grade decision support. Addressing the limitations of existing systems in open-ended consultations, Baichuan-M3 utilizes a specialized training pipeline to model the systematic workflow of a physician. Key capabilities include: (i) proactive information acquisition to resolve ambiguity; (ii) long-horizon reasoning that unifies scattered evidence into coherent diagnoses; and (iii) adaptive hallucination suppression to ensure factual reliability. Empirical evaluations demonstrate that Baichuan-M3 achieves state-of-the-art results on HealthBench, the newly introduced HealthBench-Hallu and ScanBench, significantly outperforming GPT-5.2 in clinical inquiry, advisory and safety. The models are publicly available at https://huggingface.co/collections/baichuan-inc/baichuan-m3.