SAR updates its first homegrown AI model

Hong Kong introduced HKGAI V3, an upgraded version of its first homegrown large language model, on Wednesday. Developed by the Hong Kong Generative AI Research and Development Center (HKGAI), the model improves token compression efficiency by over tenfold and extends uninterrupted agent runtime to up to 28 hours per session. The new model integrates Hong Kong’s cultural context, allowing it to better interpret local user inputs and generate culturally attuned responses. Secretary for Innovation, Technology and Industry Sun Dong emphasized that the release positions Hong Kong as a leader in AI-driven innovation, enabling practical applications like enhanced AI chatbot HKChat and document processing tool HKPilot. HKGAI V3’s Agent Workshop supports autonomous task execution, external tool integration, and cross-platform collaboration for workflow completion. The model will be open-sourced under the name ClawNet, facilitating customized AI agent development for businesses and research institutions. Guo Yike, director of HKGAI and provost of the Hong Kong University of Science and Technology, highlighted the importance of localization and security compliance over technical benchmarks. He compared pretraining a model to raising a child, stressing that alignment with local regulations and societal values is crucial for meaningful AI deployment. The government aims to advance AI industrialization and integration with traditional industries as part of broader efforts to strengthen Hong Kong’s AI ecosystem. HKGAI V1, released in February 2025, marked Hong Kong’s first homegrown large language model, and V3 builds on this foundation with improved performance and accessibility.