HONG KONG, June 16, 2026 /PRNewswire/ — Xunfei Healthcare (2506.HK) officially launched Xunfei Spark Medical Model V3.5 on June 9. The upgrade focuses on two core areas of urgent demand: clinical diagnosis and treatment, and health management for residents. According to Xunfei Healthcare, the model outperforms its Chinese competitors across key tasks including medical knowledge question answering, diagnostic and treatment recommendations, and medical document generation. It also ranked first overall in IDC’s China Medical Large Model Technology Assessment 2026 and topped the Shanghai AI Laboratory MedBench agent evaluation with a score of 98.9.
As the global development of AI healthcare shifts from a “parameter race” toward real-world deployment capability, Xunfei Healthcare’s latest upgrade demonstrates its practical value in real clinical settings through three core technological breakthroughs.
Authoritative Evaluations and Three Major Technological Breakthroughs
First, its multimodal fusion capabilities have been comprehensively upgraded. Xunfei Spark Medical Model V3.5 is among the first in the medical services sector to achieve deep integration of medical speech recognition, multi-speaker speech recognition, and automated medical record generation in real consultation-room environments. The model has been deployed in multiple leading public hospitals across China and applied in real-world clinical scenarios. The physician adoption rate for generated medical records stands at 91%, while medical record documentation time has been reduced by 52%. The model’s multimodal capabilities in imaging-assisted diagnosis and medical semantic understanding have also been significantly enhanced. Its imaging quality control has surpassed the level of expert consultation, while image recognition has been deeply integrated with diagnostic and treatment reasoning. For X-ray and MR reports, physician adoption of generated reports has reached 75%, marking the first time that imaging report generation has reached a practical level of application.
Second, efficient long-context reasoning now supports independent local deployment. Xunfei Spark Medical Model V3.5 is among the first medical models to implement efficient long-text training based on Dynamic Sparse Attention (DSA) and Multi-Token Prediction (MTP). In medical long-context reasoning scenarios, inference throughput has increased by 4.5 times. This allows hospitals and regional medical institutions to deploy and iterate models locally without relying on expensive cloud computing resources, helping the medical AI industry enter a new stage of development.
Third, an evidence-based clinical reasoning framework has significantly enhanced the model’s agent capabilities. The model integrates high-quality medical evidence, including authoritative Chinese and English clinical guidelines and expert consensus documents. It supports multi-step reasoning, deep reflection, and full evidence traceability, covering online consultations, outpatient care, and inpatient scenarios. For clinical physicians, the diagnostic and treatment assistant agent supports key functions including medical record quality control, intelligent medication review, and general and cross-specialty assisted diagnosis. These capabilities help reduce the risk of misdiagnosis and missed diagnosis while improving the standardization of medical care.
Data Flywheel Strengthens Xunfei Healthcare’s Industry Moat
Xunfei Spark Medical Model V3.5 is continuously trained on 1.6 billion anonymized medical voice records and 1.2 billion real-world clinical diagnosis and treatment records from China. Each day, the platform learns from more than 2.2 million newly added real-world medical data records, including voice, imaging, and medical record data. Through the continuous accumulation of high-quality data, Xunfei Healthcare has built a unique industry data flywheel and a self-reinforcing competitive moat. In terms of algorithmic innovation, the model uses high-quality medical data and supervised data generated from physician interaction feedback. It has developed evidence-based reasoning technology for automated training on complex tasks through a process of “evidence alignment, reflective verification, and expert reinforcement,” significantly improving the accuracy of medical foundation models.
According to IDC’s China Medical Large Model Technology Assessment 2026, Xunfei Spark Medical Model ranked first in the industry in overall strength. The assessment covered 15 core indicators, with the model ranking first in 12 capabilities, including assisted diagnosis, medical record generation, health consultation, post-diagnosis management, and follow-up care. The results show that the model demonstrates leading capabilities in product implementation, safety, and scalable application.
In the latest MedBench agent evaluation by Shanghai AI Laboratory, the model topped the ranking with an overall score of 98.9. It also ranked first in China across multiple core capabilities, including medical scenario perception and interaction, medical multi-agent collaboration, and medical safety, ethics, and compliance. OpenEvidence, a U.S.-based evidence-based medical AI tool, is reportedly valued at USD 12 billion and is already used extensively by nearly 80% of physicians in the United States. Its annual revenue in 2025 is expected to reach USD 150 million. In China, Xunfei Healthcare is developing an evidence-based diagnostic and treatment assistant agent adapted to domestic clinical standards and compatible with both Chinese and Western medical systems. At the same time, the company is expanding its global user base and increasing clinical penetration, supporting the development of a global growth curve for medical AI.
Overall, Xunfei Spark Medical Model V3.5, developed by Xunfei Healthcare Technology (2506.HK), is backed by massive volumes of real-world anonymized medical data and authoritative evaluation results. In Hong Kong’s listed AI healthcare sector, the company stands out as a rare, differentiated player with clear technological advantages. These advantages are expected to be gradually released over time. That said, the commercialization cycle for medical AI remains relatively long. Moving from technological leadership to large-scale adoption will require stronger hospital repurchase rates and higher long-term usage among new customers. For investors tracking technology-driven healthcare names, the launch of Spark Medical Foundation Model V3.5 represents an important technological milestone.