Tencent unveils Hy4 with image understanding and autonomous computer use

Tencent has released Hy4, its first multimodal language model with image understanding and the ability to operate software independently. The company says it was trained on substantially more data than its predecessor and optimized for coding, finance, office work, and professional tools including Blender and CAD applications. Model weights are planned for a later, unspecified date; meanwhile, Hy4 is available through Tencent Cloud, OpenRouter, and Tencent applications such as Yuanbao.

Hy4 can inspect detailed photographs and diagrams, generate image crops through code, and complete sequences involving several applications. Tencent says its gaming teams adapted it for Unreal, Unity, Godot, and Honor of Kings. In an internal finance evaluation covering nearly 600 tasks, it reportedly outperformed several Chinese competitors. The company also says its web-search hallucination rate fell from nearly 16% to about 5% in internal tests.

Artificial Analysis gave Hy4 35 points on its Intelligence Index, placing it second among Chinese lab models behind Kimi K3.1 Pro at 46. However, the assessment found weak factual reliability without search, while Tencent acknowledges that Hy4 trails leading commercial systems on advanced mathematics, lengthy coding projects, and complex image analysis. Input pricing is about $0.83 per million tokens and output pricing $2.50.