幻觉率不到3%，王小川把医生版的DeepSeek免费了

Core Viewpoint - Baichuan Intelligent has launched the Baichuan-M3 Plus model, marking a significant breakthrough in AI applications in the medical field, enhancing medical problem reasoning capabilities in real-world scenarios [1][3]. Group 1: Model Performance and Features - The M3 Plus model achieves unprecedented accuracy in medical scenarios, significantly improving reasoning efficiency and crossing critical thresholds for AI usability in healthcare [3][29]. - The model employs Fact-Aware Reinforcement Learning (Fact-Aware RL), which allows for the verification of AI-generated medical judgments against authoritative sources, thus reducing hallucination rates to the lowest globally [6][7]. - M3 Plus has a hallucination rate of only 2.6%, which is over 30% lower than GPT-5.2 and surpasses current industry benchmarks [10][18]. Group 2: Trust and Reliability in AI - Trust remains a significant barrier for AI adoption in healthcare, as many existing AI models produce seemingly professional but often incorrect information [5][16]. - Baichuan Intelligent has introduced "Evidence Anchoring" technology, ensuring that every medical conclusion generated by the AI can be traced back to specific evidence in original research or guidelines, achieving over 95% accuracy in citation matching [16][18]. Group 3: Accessibility and Cost - Baichuan has launched the "Haina Baichuan Plan," offering free access to the M3 Plus API for all service providers to medical professionals, aiming to promote the widespread adoption of AI tools in clinical settings [20][22]. - The cost of deploying M3 Plus for all clinical doctors in China is estimated to be around 100 million yuan annually, which Baichuan is willing to cover to foster ecosystem development [23]. Group 4: Engineering and Optimization - The M3 Plus model has undergone extensive engineering optimization, reducing API call costs by 70% compared to its predecessor while maintaining model performance and reliability [24]. - The Gated Eagle-3 decoding framework enhances prediction accuracy without significantly increasing computational costs, allowing for dynamic filtering of external information [24][25]. Group 5: Industry Context and Future Outlook - The year is seen as pivotal for AI in healthcare, with numerous significant developments indicating that the medical field is becoming a core application area for AI technology [30][32]. - Baichuan's strategy focuses on integrating AI directly into hospital core departments, positioning itself as a "second brain" for doctors rather than merely a health assistant [32][33].