UNISOUND (09678) has released its self-developed multimodal medical imaging model "U2-RadiMed".

UNISOUND (09678) announced that following the release of the U2-Med tri-medical model, the company has officially launched its self-developed medical imaging multimodal large model "U2-RadiMed." This model is the first in China to deeply integrate medical image understanding, report generation, clinical reasoning decision-making, and visual question answering into a full-link imaging large model. It supports both single and multi-image, as well as text multimodal mixed inputs, comprehensively covering mainstream medical imaging types such as CXR (chest X-ray), CT, and MRI. It establishes a core capability framework consisting of six major areas: report generation, lesion detection and localization, clinical reasoning, clinical decision-making, general medical VQA, and advanced medical reasoning, enabling a complete operational loop from "viewing images retrieving knowledge making inferences drawing conclusions." In various authoritative horizontal assessments, U2-RadiMed ranked first in comprehensive evaluation with a score of 55.47, leading domestic and international models such as Gemini 3.1 and GPT-5.4. In specialized assessments of the six core capabilities, it secured the top position in report generation, lesion detection and localization, general medical VQA, and advanced medical reasoning, with lesion detection and localization significantly outperforming similar models with a score of 37.58. Moreover, in the MedBench 5.0 full-modal large model evaluation, it achieved the highest score among participating models at 57.2. At the application level, U2-RadiMed breaks through the limitations of traditional AI in single-image analysis, capable of understanding the relational logic between multiple time-sequence images. It supports diverse clinical scenarios such as lesion detection and localization, multi-time-sequence image comparison, disease progression tracking, differential diagnosis, and report-assisted generation, providing comprehensive AI assistance for clinical physicians from lesion screening and precise diagnosis to personalized treatment decision-making. Currently, the model has been launched on the company's MaaS platform (Token Hub), supporting multi-image inputs and 40K context, allowing for immediate applicability of model capabilities and quick integration with imaging departments and clinical treatment scenarios in medical institutions at all levels. The release of U2-RadiMed signifies that the company's large model capabilities have officially entered the more complex and specialized core clinical scene of medical imaging, further refining the company's healthcare AI product system consisting of "general foundation large model tri-medical specialized large model medical imaging multimodal large model," and consolidating the company's technological leadership in the field of medical artificial intelligence. In the future, the company will continue to deepen the application and innovation of large models in the broader healthcare field, driving the transformation of medical imaging AI from technological breakthroughs to true medical productivity, assisting medical institutions in improving quality and efficiency, and creating long-term value for shareholders.