Tencent has revealed Hy4 Preview, a large scale foundation AI model built to challenge the leading players in China's artificial intelligence market. The system packs 770 billion parameters and processes up to 1 million tokens in a single context window. Internal testing indicates it edges out rival offerings from Zhipu AI and Moonshot AI.
The jump in scale represents a major technical upgrade over previous iterations. With a 1 million token context window, the model can retain vast amounts of data across single prompts. This expanded memory helps the system track complex instructions and generate detailed answers with greater precision. In internal blind engineering evaluations run by Tencent specialists, Hy4 Preview scored slightly higher than Zhipu AI with GLM 5.3 and Moonshot AI with Kimi K3.
External results show a much tighter race. In third party benchmark evaluations, Hy4 Preview traded wins with domestic rivals, including Alibaba and its Qwen 3.8 Max model. To demonstrate practical use cases, Tencent released video demonstrations showing the model handling game asset development, 3D web design, and corporate accounting compliance checks. These demonstrations highlight Tencent’s ambition to establish itself as a heavyweight in the enterprise artificial intelligence space.
