DIAGENS-B (02526.HK): iMed-6.0 Models Lead Benchmarks, Thoth Tops Overall Score

On 12 Nov 2026, DIAGENS-B (02526.HK) said its iMed-6.0 models led MMMU and MedBench tests, with Thoth first overall and Aegis second in health and medicine.

NewTimeSpace News: On 12 November 2026, Hangzhou Diagens Biotechnology Co., Ltd. (stock code: 2526) published a voluntary announcement on technical progress of its iMed-6.0 series of medical imaging large models, comprising Thoth at a medium parameter scale and Aegis at a smaller scale.

The iMed-6.0 models achieved leading global results on MMMU (Test), a large-scale multidisciplinary multimodal understanding and reasoning benchmark, and on MedBench, a medical AI evaluation platform for full-modality models. MMMU (Test) covers six domains including medicine, science and engineering across 30 categories. According to the Company's test results, Thoth ranked first overall across all domains, while Thoth and Aegis ranked first and second globally in the health and medicine category; on MedBench, their composite scores exceeded the highest scores of models then listed on the public leaderboard, ranking first and second globally.

Thoth scored 88.8 in MedBench's medical visual perception and text extraction category, 20.3% above the previous high. A general CT vertical model built on Aegis supports chest and abdominal plain and contrast CT and covers 509 diseases and imaging signs. In an evaluation using nearly 40,000 CT image-report pairs from multiple hospitals, the average AUC for 249 key tasks reached 0.930, while Aegis averaged 0.855 across all 509 tasks; the RADAR study published in Science covered 146 abdominal findings with a multi-centre average AUC of 0.895.

The Company is upgrading its in-house DoctorBench evaluation benchmark and plans to publish results for iMed-6.0 Horus, its largest model, together with development progress. The announcement noted that leaderboard comparisons were as at 11 November 2026 and do not mean the models are listed on the leaderboards; CT figures are zero-shot macro-average AUC and are not equivalent to diagnostic accuracy.

×
Share to WeChat

Open WeChat, use the "Scan", and share to my Moments.