The Zhitong Finance App learned that on October 22, the iFLY Spark Healthcare Big Model V3.7 upgrade was released. According to the company's disclosure, the new model relies on a national production computing power base to achieve a multi-modal technology breakthrough in deep integration of voice, video, and semantics. In the MedBench 5.0 agent evaluation recently released by the Shanghai Artificial Intelligence Laboratory, it reached the top with a comprehensive score of 99.6. Its performance in core tasks such as medical knowledge questioning, diagnosis and treatment recommendations, and medical document generation all surpassed mainstream models at home and abroad.
Scenario implementation capabilities have been upgraded simultaneously. In the medical record writing scenario, the reasonable rate of end-to-end model generation has increased to 85%; medical image report generation and in-depth quality control capabilities have reached industry-leading levels. For residents, iFLYX has simultaneously completed functional iterations, adding services such as multi-modal consultation of traditional Chinese medicine and dynamic memory management of health files. The results of this upgrade will fully cover the company's core business segments such as AI diagnosis and treatment assistants and AI health assistants.
The data flywheel formed by actual business accumulation has become the core driving force for model iteration. Currently, doctor-side AI assistance is used an average of 1.71 million times per day, and the average daily service on the patient side covers 520,000. The “implementation application - precipitation data - optimization model” continues to operate in a closed loop, continuously optimizing the clinical adaptation effect of the model.