Tencent’s Hy4 Can See and Use a Computer — but Its Open Weights Are Not Out Yet
Tencent released Hy4 on 4 November, the first Hunyuan model with visual understanding and computer use. It is live on TokenHub, OpenRouter and Tencent's apps — but the open weights are still to come, and its benchmark claims are Tencent's own.
Tencent released Hy4 on 4 November, and it is a genuine first for the company’s Hunyuan line: the first of its models that can understand images, and the first that can operate a computer, working through both the command line and a graphical interface, according to the Tencent announcement, 4 November 2026 and a detailed DEV Community explainer, 4 November 2026. The honest caveat belongs in the headline too: the open weights are not out yet, and the performance claims are Tencent’s own.
Highlights
- Hy4 is the first Hunyuan model with visual understanding: fine-grained image processing, chart analysis and mathematical visual reasoning.
- It is also the first with computer use, operating machines via command line and graphical interface.
- The model has 770 billion total parameters in a mixture-of-experts design; the preview architecture activates 49 billion per token.
- It is available now via Tencent Cloud TokenHub, OpenRouter and Tencent apps including WorkBuddy, CodeBuddy, Yuanbao, ima and Dayuan.
- WorkBuddy and CodeBuddy are free from 4 to 15 November — but the open weights are still to come in the coming days.
What is actually new
Through advanced visual understanding, reasoning and tool-calling, Hy4 supports fine-grained image processing, complex chart analysis and mathematical visual reasoning, Tencent said in its Tencent announcement, 4 November 2026. Its computer-use capability lets it interact with both command-line interfaces and graphical user interfaces to perform work and everyday tasks, and the model has undergone specialised training on professional software such as Blender and CAD, significantly enhancing its ability to plan and execute long-horizon tasks. Tencent also reports major improvements in instruction following, long-context comprehension and long-horizon agentic tasks. The scale is large: 770 billion total parameters in a mixture-of-experts architecture, up from Hy3’s 295 billion, with the preview architecture activating 49 billion parameters per token, as the DEV Community explainer, 4 November 2026 explains.
Where you can use it today
Hy4 is currently available across Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom’s AI assistant Dayuan, ima and more, and its API is being made available via third-party platforms such as OpenRouter, Hermes and OpenCode, according to the Tencent announcement, 4 November 2026. From 4 to 15 November, WorkBuddy and CodeBuddy will be available for free, and Hy4 is offered on TokenHub at 50% off standard pricing through 31 December 2026. Tencent lists pricing at approximately US$0.90 per million input tokens and US$2.70 per million output tokens. A lighter Hy4 Flash, focused on cost-efficiency, is scheduled for release soon.
The honest caveats
Two cautions matter. First, the open part is still a promise: “Tencent Hy plans to release model weights and system cards in the coming days,” the company said — until then, nobody outside Tencent can download, run or fine-tune Hy4 itself, as the DEV Community explainer, 4 November 2026 notes. Second, the benchmark story is company-reported, not independently verified. Tencent’s own WorkBuddy evaluation, for example, gives Hy4 an average performance score of 83.18 and a pass rate of 96.15 percent — Tencent’s figures, from Tencent’s tests. Tencent also says Hy4 is advancing into recursive self-improvement research, which is exactly the trajectory behind Anthropic’s warning that AI could soon speed up AI development beyond evaluation. The launch also sharpens China’s open-weight race, in a market where even US firms are localising fast, as Apple Intelligence’s planned China launch with iOS 27.3 shows. What decides Hy4’s reputation is not this announcement but the download, and whether the capabilities hold up under outside testing.
