Tencent has released Hy4, a model that adds image understanding and computer-use capabilities to its language-model line. Hy4 is currently available through Tencent Cloud, OpenRouter, and Tencent applications including Yuanbao. Tencent says the model weights will be released later, but it has not provided a date.
The model was trained for professional tasks involving coding, finance, and office software. Tencent demonstrated Hy4 locating a roughly two-meter object in a 2.8-gigapixel aerial image by generating cropped sections through code. It can also control tools such as Blender and CAD software and handle workflows that cross multiple applications. For game development, Tencent worked with its studios to adapt the model for Unreal, Unity, and Godot.
Tencent reports that Hy4 performed well in an evaluation involving more than 200 finance professionals and nearly 600 tasks. The company also says a revised search system reduced hallucinations in its internal web-research tests from nearly 16 percent to about 5 percent. Artificial Analysis gave Hy4 35 points on its Intelligence Index, tying Grok 4.7 and placing it second among Chinese lab models behind Kimi K3.1 Pro.
The model still has stated weaknesses. Without web search, it answered about half of questions correctly on an Artificial Analysis factual-accuracy test and invented answers 89 percent of the time when it lacked the information. Tencent also acknowledges gaps on advanced mathematics, long coding projects, and complex image analysis.