The Hong Kong Generative AI Research and Development Centre launched HKGAI-V3 on June 3.

The DeepSeek-based large language model is designed to run on domestic chips as the lab seeks to commercialize its products and export Chinese AI overseas.

HKGAI said the model, built on DeepSeek V4, improved token compression efficiency by more than 10x.

The launch builds on the lab’s public-sector work.

Hong Kong’s government said in April that HKPilot, its document-processing copilot, began pilot use in government bureaux and departments from mid-2024, while HKChat was being prepared for public release.

🔗 Source: South China Morning Post