Tencent unleashes 770B-parameter Hy4
Tencent has unveiled Hy4, a new 770-billion-parameter AI model, now available in a preview release. The model is designed to support coding, general office tasks, and scientific research applications. A key feature of Hy4 is its extensive context window, which is reported to exceed 1 million tokens. Tencent has also open-sourced this advanced AI model, making its capabilities accessible to a broader developer community and potentially accelerating its adoption across various industries.
Tencent’s open-sourcing of the 770-billion-parameter Hy4 model with a 1-million-token context window is a significant development for Asia’s AI landscape. This move directly challenges the dominance of Western models in enterprise applications, particularly in coding and scientific research. By making such a large model openly available, Tencent aims to foster a developer ecosystem around its technology, potentially accelerating AI adoption in Chinese and regional businesses. The real story here is not just the model's size, but its strategic positioning. Tencent is betting on open access to drive integration into existing enterprise workflows across Asia. The risk, however, lies in the quality of community contributions and the speed of enterprise adoption compared to established proprietary solutions. Our view is that this could significantly boost local AI development, but its long-term impact hinges on practical, real-world performance and developer engagement beyond the initial release.
Related reading
6 stories
How AI tools are reshaping stock trading, investment research in China
Tencent's Hy4 model aims to reshape enterprise AI, much like how AI tools are transforming finance in China.

People’s Daily rejects US claims of malicious AI distillation, warns of countermeasures

Databricks to Invest Over US$350 Million in Singapore, Double Its Workforce

Personetics Launches AI Banking Console for Relationship Managers

Anthropic Sets October Launch for Singapore Office With Local Hiring Planned

