BaiChuan's M2Plus Model Cuts Medical Hallucinations by 3x
BaiChuan Unveils Breakthrough Medical AI Model M2Plus
Beijing, October 22, 2025 – AI company BaiChuan has launched its next-generation medical large language model M2Plus, achieving what industry experts call "the most significant reduction in medical hallucination rates" to date. The release coincides with upgrades to its companion application BaiXiaoYing and new API availability.
Superior Performance Metrics
Independent evaluations show M2Plus reduces medical hallucinations (incorrect or fabricated information) by approximately three times compared to existing models like DeepSeek. Notably, it outperforms even OpenEvidence, currently the leading US-based medical AI application.
Image source: AI-generated via Midjourney licensing
Six-Source Evidence Framework
The model's breakthrough stems from its proprietary six-source Evidence-Based Reasoning (EAR) architecture:
- Original clinical research
- Systematic evidence reviews
- Medical guidelines
- Practical clinical knowledge
- Public health education materials
- Regulatory/real-world data
"This isn't just another chatbot with a medical vocabulary," explains Dr. Li Wei, BaiChuan's Chief Medical Officer. "M2Plus functions as a clinical decision support system, cross-referencing multiple authoritative sources before generating responses."
Technical Innovations
The model employs:
- PICO framework: Transforms natural language queries into structured clinical questions for precise evidence retrieval
- Evidence-enhanced training: Requires responses to cite verifiable sources while suppressing speculative content
- Dynamic quality scoring: Rates evidence reliability using criteria similar to peer-review standards
Real-World Validation
In clinical testing across 12 hospitals:
- 94% accuracy in diagnostic suggestions (vs. 78% for previous models)
- Reduced physician verification time by 62%
- Scored 97/100 on the USMLE licensing exam – surpassing most human test-takers
"The difference is night and day," reports thoracic surgeon Dr. Zhang Yu from Peking Union Medical College Hospital. "When discussing rare oncology cases, M2Plus provides properly caveated answers with clear evidence hierarchies."
The company cautions that M2Plus remains an assistive tool rather than autonomous diagnostician, with all outputs requiring clinician review.
Key Points:
✅ Hallucination reduction: Cuts incorrect outputs by ~3x versus competitors ✅ Clinical validation: Achieves near-perfect USMLE exam score (97/100) ✅ Multi-source verification: Integrates six evidence types via EAR framework ✅ Physician adoption: Reduces information verification time by over 60%