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

2026-02-06

AI Summary

This work addresses the limitations of existing large medical language models in open-ended clinical consultations, where they often lack proactive information gathering, systematic reasoning, and robust factual reliability—hindering their utility for clinical-grade decision support. To bridge this gap, we propose Baichuan-M3, a medically augmented large language model that unifies active questioning, multi-turn evidence-integrated reasoning, and adaptive hallucination suppression within a single framework, effectively emulating the structured diagnostic workflow of expert clinicians. Through a specialized training pipeline, Baichuan-M3 achieves state-of-the-art performance on HealthBench, HealthBench-Hallu, and ScanBench, demonstrating significantly superior clinical consultation quality, recommendation reasonableness, and safety compared to GPT-5.2. This advancement marks a pivotal shift in medical AI from passive question-answering toward active, clinically grounded reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• clinical inquiry
• medical decision-making
• hallucination suppression
• open-ended consultation
• large language model

Innovation

• clinical inquiry modeling
• active decision support
• long-horizon reasoning
• adaptive hallucination suppression
• medical large language model