DIAGENS-B(02526) iMed-6.0 Series Medical Imaging Large Models Achieve Significant Technical Progress, Attaining World-Leading Results on MMMU (Test)

DIAGENS-B(02526) announced that the Group has continued to deepen its work in medical imaging artificial intelligence and independently developed the iMedImage foundation model. In the first half of 2026, it completed a full-modality architecture upgrade, and in July of the same year launched the iMedLoop medical imaging AI research and production acceleration platform. Leveraging the Group's accumulated large-scale, high-quality medical imaging data, through underlying algorithm technology innovation, continuous training, and expanded training scale, the medical imaging understanding and analysis capabilities of the Group's foundation models have continued to improve. The board of directors is pleased to announce that the Group's latest-generation medical imaging large models, iMed-6.0 Thoth (Thoth) and iMed-6.0 Aegis (Aegis), have achieved significant technical progress. These are respectively the medium-parameter-scale and smaller-parameter-scale versions in the iMed-6.0 series. The Group's iMed-6.0 series models achieved world-leading results on the globally renowned large-scale multidisciplinary multimodal understanding and reasoning benchmark (test set) ("MMMU (Test)") and in the full-modality large model evaluation on MedBench, a professional evaluation platform for medical AI. MMMU (Test) covers six major fields including medicine, science, and engineering, across 30 categories. According to the Group's test results, Thoth ranked first in the overall score across all fields, higher than other models, while Thoth and Aegis ranked first and second globally in the "Health and Medicine" subcategory, demonstrating their outstanding medical multimodal understanding and reasoning capabilities and further reflecting their cross-disciplinary, multi-task generalization potential. According to the evaluation results obtained after the Group's models were submitted to MedBench, the overall scores of Thoth and Aegis were also both higher than the highest scores among full-modality large models listed on the public leaderboard during the same period, ranking first and second globally. Among them, Thoth scored 88.8 in the "Medical Visual Perception and Text Extraction" subcategory, exceeding the previous highest score by 20.3%. The Group's general computed tomography (CT) vertical model developed based on Aegis has achieved important technical progress in the comprehensive analysis of multiple organs and multiple abnormalities in CT images. This model supports non-contrast and contrast-enhanced CT of the chest and abdomen, and can perform comprehensive analysis of multiple organs and multiple types of abnormalities within the scan range of a single examination, covering 509 diseases and imaging signs. In the Group's evaluation using nearly 40,000 pairs of CT images and reports from multiple hospitals, the average area under the curve (AUC) (an indicator of abnormality discrimination ability, with values closer to 1 being better) for the 249 key tasks selected by the Group reached 0.930. As a public research reference, the RADAR study published in Science disclosed coverage of 146 abdominal imaging findings, with an average AUC of 0.895 validated across multiple centers. Aegis's average AUC across all 509 tasks was 0.855. To further enhance core technical capabilities, the Group firmly believes that building a more objective, comprehensive, clinically relevant, and high-difficulty professional evaluation benchmark is crucial. The Group is continuously upgrading its self-built DoctorBench professional evaluation benchmark to further accurately and comprehensively measure models' medical imaging understanding, reasoning, and integrated application capabilities in complex scenarios using challenging medical tasks. The Group plans to announce a new round of evaluation results subsequently, including the relevant performance and R&D progress of the Group's largest-parameter model, iMed-6.0 Horus. The board of directors believes that the continuous improvement of foundation model capabilities is an important DRIVE for enhancing the core competitiveness of the Group's "model factory" and expanding commercialization space. Stronger foundation models are expected to improve the model development and training efficiency of iMedLoop, accelerate the supply of specialized models for different organs, diseases, and application scenarios, and support medical experts in accelerating research and validation of new methods. At the same time, the relevant technical progress is expected to accelerate the deployment and application of diagnostic models in primary healthcare systems, help medical institutions improve the efficiency and accuracy of imaging reports through inference services, and drive the expansion of the case-based fee model inference service business. As the number of available models and service scenarios increases, the Group is expected to expand customer coverage and case service scale, driving the transformation of technological advantages into sustained commercial value. The Group will continue to explore the upper limits of foundation model capabilities, strengthen training and professional evaluation, and is committed to accelerating the intelligentization of global medical imaging and improving human health.