Tencent releases Hy4, an AI model that can use a computer, and says it left hacking material out of training

Tencent, the Chinese company behind the WeChat messaging app, released a new AI model called Hy4 on Wednesday.  It said the model ranks among the world's best "open-source" models.

Open-source, or open-weight, means the company publishes the model's weights, the billions of numbers that make up its trained "brain," so anyone can download and run it on their own computers.  The models behind ChatGPT and Claude are not shared that way.  Tencent said it will post Hy4's weights and system cards, documents describing how the model was built and tested, in the coming days.

Tencent said Hy4 is its first model that can understand images and that can operate a computer.  That means it can type commands and click through programs on a screen, much like a person would.  Tencent said it trained Hy4 on professional software such as the 3D design program Blender and computer-aided design, or CAD, tools.

The company also highlighted safety.  It said Hy4 was trained on defensive cybersecurity material, such as fixing software flaws, but that material about exploiting flaws to break into systems was left out.  Tencent said experts at its own Xuanwu Lab and Keen Security Lab tested the model and found no significant concerns about its hacking abilities in the areas they checked.  It said Hy4's ability to design open-ended biology experiments is more limited than leading models.  Those are the company's own findings.

Tencent said Hy4 cut its rate of hallucinations, made-up answers, to 7.2% from 12.95% in its own tests.  In a chart it published, Hy4 scored below Anthropic's Claude Opus 5.5 and OpenAI's GPT 6.1 Sol on most of the tests shown.

Hy4 costs about 90 cents per million tokens of input, the chunks of text a model reads, Tencent said.  It is available in Tencent apps and through outside services such as OpenRouter.

Source:  Tencent announcement, Nov. 4, 2026