Tencent has released Hy4, the full version of its flagship Hunyuan artificial intelligence model, giving the system two abilities it never had before: it can see images and it can operate a computer. Announced on Wednesday, November 4, it moves Tencent's top model into the front rank of China's open-model race — with one catch. The model is available to use now, but the open weights that would let anyone download and run it have not been published yet.
Hy4 is a 770-billion-parameter mixture-of-experts system, up from Hy3's 295 billion parameters. It succeeds a preview version shown in August and is live across Tencent's own products, including the Yuanbao chatbot, WorkBuddy, CodeBuddy, the ima app and the WeCom assistant Dayuan, as well as on Tencent Cloud's TokenHub. Developers outside China can also reach it through API platforms including OpenRouter, which lists a context window of about one million tokens.
Vision and computer use arrive together
The headline change is vision. For the first time in the Hunyuan line, Hy4 can process images in fine detail, analyse complex charts and work through mathematical problems presented visually, combining what it sees with reasoning and tool use. Alongside that comes computer use: the model can carry out multi-step tasks by working through a graphical interface and the command line, rather than only answering in text.
Tencent says the combination lifts Hy4 into the ranks of the world's strongest open-source models. In a financial test run by the company with 213 experts across 579 real tasks, Hy4 scored 3.04, ahead of Kimi K3.1 Pro at 2.68, GLM-5.3 at 2.68 and DeepSeek-V4.1-Flash at 2.97. That test was organised by Tencent itself, so independent benchmarks will matter more — but the model is clearly built for practical office, finance, development and research work, not just chat.
Open in name, hosted for now
The distinction matters: what users can access today is the hosted model, not the model itself. Tencent says it plans to release the weights in the coming days, alongside a fuller system card covering safety results, with support from the vLLM and SGLang projects. Until that release lands, nobody outside Tencent can download, fine-tune or inspect Hy4 directly. A lighter version, Hy4 Flash, focused on speed and lower cost, is also coming soon, with no firm date yet.
Pricing is unchanged from the preview: 6 yuan per million input tokens, 18 yuan per million output tokens and 0.3 yuan per million cached tokens — roughly $0.90, $2.70 and $0.045 respectively. To gather feedback, Tencent is offering free use in WorkBuddy and CodeBuddy from November 4 to 15, and a 50 percent discount on TokenHub pricing through December 31, 2026.
A safety-first release in a nervous moment
The launch also carries a deliberate safety message. Tencent said Hy4 was trained with limited cybersecurity data, filtering out material such as exploit development, reverse engineering and penetration testing to reduce the risk of the model helping with cyberattacks. The company said it recognises the potential safety risks posed by advanced cyber and biological capabilities and is committed to a more responsible approach to training, open-sourcing and release.
That caution follows a similar move by Moonshot AI in October and comes after a wave of AI-driven cyberattacks raised concern about what powerful models can do in the wrong hands. What happens next is a short checklist: the weights, the Flash version and independent test results. If the weights arrive as promised, Hy4 becomes a genuine open-weight contender; until then, it is a strong hosted model with an open promise attached.



