Table of Contents
On November 4, 2026, Tencent officially released the full version of Tencent Hy4, its new flagship AI model. The launch adds two major abilities the earlier preview did not have: native visual understanding and computer use. In plain terms, Tencent Hy4 can now look at images and charts, and it can operate software on your behalf. For anyone who follows AI news, this is one of the most important releases of the month, and it arrives in the middle of a remarkable burst of new models from labs around the world.
Tencent Hy4 is the latest model from Tencent’s Hunyuan team, the group behind the company’s large language models. The full release builds on the Hy4 preview that appeared in August 2026, but it is a much bigger step: Tencent says it substantially scaled up both pre-training and post-training, and it reports gains over the preview in instruction following, long-context understanding, long-horizon tasks, and deep research. Here are seven powerful facts you should know about it.
1. Tencent Hy4 Brings Vision to the Hunyuan Line
For the first time in the Hunyuan line, Tencent Hy4 ships with native visual understanding. That means the model can interpret fine-grained image details and complex charts, not just text. This matters for real work: a model that can read a chart, a screenshot, or a diagram can help with reports, research, and analysis in ways a text-only model cannot. Vision turns Tencent Hy4 from a chat tool into a colleague that can actually see what you are working on.
2. Computer Use Lets Tencent Hy4 Operate Software
The second headline ability is computer use. Tencent Hy4 can operate across command-line and graphical interfaces, which means it can navigate software, run commands, and carry out multi-step tasks inside applications. This is the same direction the whole industry is moving in: models that do not just answer questions but take actions. Combined with vision, computer use lets Tencent Hy4 understand a screen and then act on it, a combination that makes it genuinely useful for productivity workflows.
3. Massive Scale: 770 Billion Parameters
Tencent Hy4 is a mixture-of-experts model with 770 billion total parameters, of which 49 billion are active for any given task. The mixture-of-experts design is the key idea: instead of using the whole model for every word, it routes each task to a subset of experts. That keeps the model powerful while keeping running costs manageable. The scale puts Tencent Hy4 among the largest models available, and it explains why Tencent positions it as a flagship rather than a mid-range release.
4. A 1M-Token Context Window for Long Work
Tencent Hy4 supports a context window of one million tokens. In practice, that means you can feed it very long documents, large codebases, or entire research collections and ask questions across all of them at once. Long-context understanding is one of the areas where Tencent reports the biggest gains over the preview. For deep research, legal review, or studying a long report, a million-token window removes the need to split your work into small pieces.
5. Where You Can Try Tencent Hy4 Today
Tencent Hy4 is available through hosted APIs and inside Tencent products. You can access it through Tencent Cloud and OpenRouter, and inside CodeBuddy, WorkBuddy, Yuanbao, and ima. That spread matters: developers can call the model through an API, while everyday users can try it inside productivity apps they already use. Tencent has also said that the model weights and a comprehensive system card are planned for a later release, so full open deployment is still to come. A separate Hy4 Flash release is planned for the near future, targeting better cost efficiency and faster inference.
6. What Tencent Hy4 Costs to Run
Listed API pricing is unchanged from the preview: $0.834 per million input tokens, $2.501 per million output tokens, and $0.042 per million cached input tokens. Tencent is also offering a promotional discount on its TokenHub platform through the end of 2026, which makes trying the model noticeably cheaper during the launch window. For comparison, flagship models from other labs often cost several times more per million tokens, so Tencent Hy4 is priced to compete hard for developer attention.
7. What Is Still Coming: Open Weights and Hy4 Flash
Two things are still on the way. First, the open weights: the earlier preview was open-sourced, and Tencent says the full model’s weights will follow, accompanied by a system card covering safety evaluations and technical details, with ecosystem support developed alongside the vLLM and SGLang communities. Second, Hy4 Flash, a faster and cheaper variant. Until those arrive, the practical way to use the model is through the hosted APIs and Tencent’s own products. That is worth knowing before you plan a project around self-hosting it.
Hy4 and the November 2026 Model Rush
Hy4 did not arrive alone. November 2026 has already seen a cluster of major releases, including Google’s Gemini 4 Carbon, which targets cybersecurity work. If you want to compare the two launches, read our Gemini 4 Carbon breakdown alongside this article.
The wider trend is clear: labs are racing to ship models that act, not just answer. OpenAI’s recent work on decision-making APIs points the same way, as covered in our OpenAI Decisions API explainer. And if you want to put a powerful model to creative use, our guide to creating AI product demo videos that sell shows what modern models can do for marketing. For the official word on this release, see Tencent’s announcement about its enterprise AI portfolio and the upcoming Hy4 general release.
The honest caveat: the benchmark gains quoted here are vendor-reported, not independently verified. That is normal for a launch-day announcement, and independent testing will follow. But the shape of the release is clear either way. Hy4 is a big, multimodal, action-taking model, priced to compete, and arriving at a moment when AI agents are moving from demos into daily work. Whether you are a developer, a creator, or simply curious about where AI is heading, it is a release worth watching.
How to Get Started With the New Model
If you want to try the model this week, start small. Pick one real task, such as summarizing a long report or drafting a product description, and run it through one of the hosted products first. That gives you a feel for the quality and the speed before you spend anything on API calls. Next, test the vision ability: upload a chart or a screenshot and ask questions about specific numbers or details. Vision models can be impressive, but they can also misread fine print, so check the answers against the source before you rely on them.
Finally, if you plan to build with computer use, test it in a safe environment first. Give the agent a limited task, watch what it does, and only then widen its permissions. Agents that operate software are powerful precisely because they act, so a careful rollout protects your files and accounts.
For developers, the pricing structure rewards caching: repeated context costs far less than fresh input, so design prompts that reuse context where possible. For creators and small businesses, the hosted products are the easier path. You get the model’s abilities without managing keys, quotas, or infrastructure, and the launch discount makes the trial period cheap. Either way, treat launch week as a test phase. Benchmarks will be independently checked, the open weights will arrive later, and the Flash variant may change the value equation. Starting now still makes sense, because the skills you build with vision and agent workflows transfer to whatever comes next.
Frequently Asked Questions About Hy4
What is Hy4?
Hy4 is Tencent’s flagship AI model, released in full on November 4, 2026. It is a 770-billion-parameter mixture-of-experts model with native vision and computer-use abilities, plus a one-million-token context window.
How is Hy4 different from the Hy4 preview?
The August 2026 preview was text-only and open-sourced. The full Hy4 release adds visual understanding and computer use, with substantially scaled training and reported gains in instruction following, long-context understanding, and deep research.
Can I use Hy4 for free?
Hy4 is available through hosted APIs and products such as WorkBuddy, CodeBuddy, Yuanbao, and ima, with a promotional discount on Tencent Cloud TokenHub through the end of 2026. Check the current pricing page before starting a large project, since launch offers change.
Is Hy4 open source?
Not yet for the full release. The preview was open-sourced, and Tencent says the full model’s weights and a comprehensive system card are planned for a later release. Until then, access is through hosted APIs and Tencent products.


Pingback: Inkling 2 Small: 7 Stunning Facts You Must Know