The Hong Kong Generative AI Research and Development Center (HKGAI) officially launched the HKGAI-V3 model.
HKGAI-V3 is built on the DeepSeek V4 foundation and is geared toward AI commercialization.
The model is designed to operate on Chinese domestic chip lines.
HKGAI stated that the system achieved significant improvements in performance and Agent capabilities.
Upgraded token compression technology increases processing efficiency by more than 10 times.
Reducing the number of tokens required for processing helps shorten the model’s response time.
The continuous operating time of the Agent increased nearly 100 times compared to before.
The Agent Workshop platform, developed based on HKGAI-V3, can operate continuously for up to 28 hours.
In a single session, Agent Workshop generates a complete research report.
Professor Guo Yike, Director of HKGAI and Vice-President of the Hong Kong University of Science and Technology, announced this result.
HKGAI was established in October 2023 under the Hong Kong government’s InnoHK funding program.
The center focuses on developing AI models tailored to Hong Kong’s specific needs.
HKGAI currently operates HKChat, the city’s first chatbot that supports Cantonese.
HKChat is trained to provide information on public services, local laws, and regulations.
HKGAI also develops various AI tools to support productivity for government agencies.
📌 Conclusion: HKGAI-V3 marks a new step forward for Hong Kong’s AI ecosystem as it is built on DeepSeek V4 but capable of operating on Chinese domestic chips. The center noted that token compression efficiency increased more than 10 times and the continuous operating time of the Agent increased nearly 100 times. The clearest evidence is that Agent Workshop can run stably for up to 28 hours to complete a research report. The project reflects the ambition to commercialize AI, expand the export of Chinese AI technology, and build products that serve local needs in Hong Kong.
