The iFlytek Xinghuo Medical Large Model released version V3.7 on October 22. According to reports, this version relies on a domestically produced computing power foundation and integrates voice, imaging, and semantic capabilities into a single multimodal framework. In the latest MedBench5.0 intelligent agent evaluation by the Shanghai Artificial Intelligence Laboratory, it ranked first with a comprehensive score of 99.6 points. Its performance in core tasks such as medical knowledge Q&A, diagnosis and treatment recommendations, and medical document generation surpassed those of mainstream large models both domestically and internationally.
The ability to implement solutions rises along with water levels. In the scenario that most tests whether the medical records are written like those by a real doctor, the end-to-end generation accuracy of the model reaches 85%; the generation and deep quality control of medical imaging reports also reach industry-leading levels. Xunfei Xiaoyi has been updated for ordinary residents, with new services such as traditional Chinese medicine multi-modal consultation and dynamic health record management added. The results of this upgrade will be fully integrated into Xunfei’s core business areas, including AI diagnosis assistant and AI health helper.
What actually pushes the model forward is the data wheel that has been accumulated in real business operations. Currently, the daily average calls of AI assistance on the doctor side have reached 1.71 million, and the daily service coverage for patients is 520,000 people—the closed loop of “implementation, data accumulation, and model optimization” continues to operate, gradually refining the model’s adaptation to real clinical scenarios. When the domestic computing power supports the medical large model to achieve full scores on tests and handle millions of real calls per day, it gets one step closer to truly reducing the burden on clinics.