Shanzhi-M1 is a 32 billion parameter medical large language model developed by Shanghai Mingpin Medical Data Technology Co., Ltd., built upon the Qwen3-32B base architecture. It features an innovative medical LLM alignment framework that integrates authoritative medical standards into its training pipeline, utilizing a 3D Medical Standard System and a multi-dimensional reward model. This model excels in clinical cognition and dynamic medical standards, achieving leading scores on HealthBench, and is primarily designed for medical education, clinical decision support, and public health consultation.
Loading preview...
Shanzhi-M1: A Medical LLM Alignment Framework
Shanzhi-M1 is a 32 billion parameter medical large language model developed by Shanghai Mingpin Medical Data Technology Co., Ltd., based on the Qwen3-32B architecture. It introduces a novel alignment framework designed to address common challenges in medical LLMs, such as misalignment with clinical cognition and poor adaptation to evolving medical standards.
Key Innovations & Capabilities
- 3D Medical Standard System: Integrates domain-specific standards (e.g., accuracy, compliance, empathy) into data generation, SFT, and RL to ensure clinical relevance and trustworthiness.
- Independent Multi-Dimensional Reward Model: Replaces costly real-time expert scoring with internalized, decomposed medical evaluation criteria, reducing expert labor by over 90%.
- Geometric Projection Reference Constraints: Mathematically regularizes medical cognitive logic to align scoring gradients with clinical reasoning, enabling effective training on large-scale synthetic data.
- Top-Performing Medical Model: Achieves leading scores on HealthBench (62.7 Full, 44.7 Hard), outperforming all open-source and many closed-source medical models.
- Clinical Scenario Excellence: Demonstrates high performance across 5 core medical scenarios, including Emergency Referrals (74.3) and Medical Communication (69.6).
- Cost-Efficient & Standard-Extensible: Significantly reduces annotation costs and supports dynamic updates to multi-source medical guidelines.
Ideal Use Cases
Shanzhi-M1 is intended for:
- Medical student training: For case simulations and knowledge verification.
- Healthcare provider decision support: Offering second opinions and ensuring guideline alignment.
- Public health education: Providing general health consultations.
Note: This model is for research, education, and decision support only and does not replace professional medical advice or diagnosis. Always validate outputs against authoritative medical guidelines.