Shanghai Institute Releases 'Shenzhen', an 11B Multimodal AI Model for Science
The Shanghai Institute for Advanced Study of Science has open-sourced 'Shenzhen' (Monkey King Bang/MKB), an 11-billion-parameter multimodal AI model. The model is designed to process and generate data across six scientific domains: DNA, RNA, proteins, small molecules, weather, and medical imaging. Unlike traditional specialized scientific AI models, 'Shenzhen' unifies diverse scientific modalities into a single architecture, demonstrating that a single model can learn shared patterns across different scientific disciplines. This release advances the 'AI for Science' paradigm by providing open-source weights and code to facilitate collaborative research. Built on the Qwen3-VL-8B backbone, the model preserves domain-specific data structures (like molecular connections and spatial weather distributions) before processing them through shared weights. In evaluations, it outperformed the 1-trillion-parameter Intern-S1-Pro in half of the biological sequence tasks and achieved a 91.20 average Dice score in medical image segmentation.
## BACKGROUND
"AI for Science" uses machine learning to solve complex scientific problems, but models are typically specialized for single tasks like weather forecasting or protein folding. SMILES (Simplified Molecular Input Line Entry System) is a standard chemical notation that translates 3D molecular structures into compact, computer-readable ASCII text strings.