Tencent has released an artificial intelligence model it says outperforms flagship models from Z.ai and Moonshot in tests, as the Chinese tech giant claims to have joined the country’s top artificial intelligence companies.
Image source: Tencent.com
Model Tencent Hy4 preview version With 770 billion parameters, its context window is 1 million tokens – indicating improvements in size and data handling capabilities compared to its predecessor. The larger the context window, the more information the model can remember and the better it can respond to user requests and instructions.
In a blind test conducted by the company’s own experts, the Hy4 Preview narrowly beat the Z.ai GLM-5.3 and Moonshot Kimi K3. In third-party benchmarks, its results compare differently to Chinese rivals including Alibaba’s flagship Qwen 3.8 Max. Tencent demonstrated the performance of this model in scenarios such as game development, 3D web design, and solving accounting problems.
Developers have released the model’s weights, which can be accessed through the company’s cloud platform for integrating artificial intelligence agents and chatbots into applications. Tencent promises to improve the performance of future versions of Hy4 through pre-training and post-training mechanisms.
The Chinese tech giant is considered a laggard in the artificial intelligence race – funding its research and equipment in the field through its WeChat platform and gaming projects. As of the end of last quarter, Tencent’s capital expenditures on artificial intelligence projects increased by 176% to 52.8 billion yuan ($7.9 billion). These projects are already starting to pay off: Tencent WorkBuddy has become China’s most popular desktop AI agent, sparking ByteDance and Alibaba’s counterattack in the enterprise AI market.
If you find an error, select it with your mouse and press CTRL+ENTER.