Baichuan-M2-32B: Medical-Enhanced Reasoning Model

Baichuan-M2-32B is Baichuan AI's second medical model, designed to excel in real-world medical reasoning tasks. Built upon the Qwen2.5-32B architecture, it introduces a novel Large Verifier System and advanced medical domain adaptation techniques to achieve breakthrough medical performance while preserving general capabilities.

Key Innovations & Capabilities

Large Verifier System: Incorporates a comprehensive medical verification framework with patient simulators and multi-dimensional verification (8 dimensions including medical accuracy and completeness).
Medical Domain Adaptation: Utilizes Mid-Training for efficient medical knowledge injection and a multi-stage reinforcement learning strategy to enhance medical knowledge, reasoning, and patient interaction.
Doctor-Thinking Alignment: Trained on real clinical cases and patient simulators, fostering clinical diagnostic thinking and robust patient interaction.
Leading Medical Performance: Achieves the world's leading open-source medical model status, outperforming all open-source and many proprietary models on HealthBench, with medical capabilities closest to GPT-5.
Efficient Deployment: Supports 4-bit quantization, enabling deployment on a single RTX4090, and offers 58.5% higher token throughput in its MTP version for single-user scenarios.

Performance Highlights

Baichuan-M2-32B demonstrates superior performance on medical benchmarks like HealthBench, scoring 60.1 overall, 34.7 on HealthBench-Hard, and 91.5 on HealthBench-Consensus, surpassing models like gpt-oss-120b and Qwen3-235B-A22B-Thinking-2507. It also shows strong general performance, outperforming Qwen3-32B (Thinking) on benchmarks such as AIME24, Arena-Hard-v2.0, CFBench, and WritingBench.

Intended Use Cases

Medical education
Health consultation
Clinical decision support

Note: This model is for research and reference only and cannot replace professional medical diagnosis or treatment. It is recommended for use under the guidance of medical professionals.