Tencent has released Hy4, its latest large language model, pitching it as a stronger open-source rival that can see, reason and operate software — while trying to blunt fears that powerful AI could help cyberattacks.
Key Points:
- Tencent released Hy4 on Wednesday, 4 November 2026, as its latest-generation large language model.
- The model is open source, with its weights and system cards due for release in the coming days.
- Hy4 adds visual understanding and computer-use skills for the first time in the Hy series.
- It can work through command-line and graphical interfaces, including professional tools such as Blender and CAD software.
- Tencent says it trained the model mainly on defensive cybersecurity material and excluded offensive exploitation material.
- Hy4 is available now through Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao and other Tencent products.
Tencent released Hy4 on Wednesday, its newest open-source artificial intelligence model, adding vision and computer-use abilities to its flagship system.
The launch lands as AI companies race to turn chatbots into agents that can actually do work — drafting documents, analysing charts, building software and running long tasks without constant prompting. Tencent is betting that openness will help it here: by releasing the model for others to use and build on, it hopes to spread Hy4 through workplaces, developer tools and its own apps. At the same time, the company is trying to answer a sharper question now facing the whole industry: how much dangerous capability should a public model carry? Its answer, at least for Hy4, is to draw a line in training — defensive security work in, offensive hacking material out.
What happened
Tencent announced Hy4 from Shenzhen on 4 November, describing it as a major step up from Hy4 Preview, the version that came before it. The company said the new model performs better across a wide range of uses, with particular gains in following instructions, understanding long documents and carrying out long, multi-step tasks.
Two new abilities headline the release. The first is visual understanding: Hy4 can now work with images in detail, including fine-grained image processing, complex chart analysis and mathematical visual reasoning, according to Tencent. The second is computer use. For the first time in this model family, Hy4 can interact with both command-line interfaces and graphical user interfaces — the text commands and on-screen controls people normally use — to carry out work and everyday tasks. Tencent says it gave the model specialised training on professional software such as Blender and CAD tools, aiming to help it plan and finish longer, more complicated jobs.
The model is also aimed squarely at office and professional work. On Tencent WorkBuddy, its workplace assistant, Hy4 is meant to help with visual creative tasks, 3D scene building and web development, as well as document presentation and collaboration between people and AI. Linked through WorkBuddy to WeCom, Tencent’s workplace messaging service, it can draw on chats, emails, documents, calendars and meetings to track projects, flag outstanding tasks and warn about approaching deadlines. Tencent has also tuned Hy4 for financial analysis, training it on thousands of specialised workspaces and real-world tasks in fund research, securities analysis and quantitative investment work.
Hy4 is available now across Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom’s AI assistant and ima, with its programming interface also offered through third-party platforms such as OpenRouter. Tencent says integration is under way for other products, including QQ and Tencent News. A cheaper version, Hy4 Flash, is due soon for users who prioritise cost over top performance. Pricing for Hy4 itself is unchanged from Hy4 Preview, at about $0.90 per million input tokens and $2.70 per million output tokens, with cached input tokens at about $0.045 per million. WorkBuddy and CodeBuddy are free from 4 to 15 November, and Hy4 on TokenHub is half price through 31 December 2026.
Why it matters
Open models have become a central front in the AI race, and China is a major part of that story. The same day Apple began testing its own AI tools in China, Tencent’s launch underlined how quickly Chinese companies are pushing capable models into everyday products. Readers can follow that wider shift in our coverage of Apple Intelligence’s arrival in China, where a local partner model now sits inside parts of the iPhone experience.
The safety choices around Hy4 may matter as much as its speed. The rollout comes after a series of AI-driven cyberattacks raised concern about what advanced models can do for attackers, Reuters reported. Tencent said it filtered training material such as exploit development, reverse engineering and penetration testing to limit the model’s offensive potential, and that its security teams found no significant concerns about offensive cybersecurity capabilities in the areas they assessed. “We recognize the potential safety risks posed by advanced cyber and biological capabilities, and we are committed to a more responsible approach to model training, open-sourcing, and release,” Tencent said in a statement. The company also stressed biosafety, saying internal reviews found strong basic biological knowledge but more limited ability in open-ended experimental design than frontier models.
That framing echoes a broader industry move: Chinese AI firm Moonshot AI last month limited its newest model’s cyber-offensive capabilities, Reuters reported, as Chinese and American companies alike try to prevent their models from becoming tools for attacks. Whether Hy4’s filters hold up in practice will be tested quickly, because open weights are hard to take back once they spread.
For ordinary users, the more immediate test is practical. AI tools are already spreading into classrooms, clinics and offices, as our earlier report on new AI tools at work in classrooms and clinics showed — and Hy4’s pitch is that it can stay useful across a whole working day, not just a single prompt. If it does, it could become the engine behind more of Tencent’s apps, from workplace software to games, where Tencent says Hy4 has improved its handling of complex game logic through integration with engines including Unreal Engine, Unity and Godot.
What happens next
Watch three things. First, the open release itself: Tencent plans to publish the model weights and system cards in the coming days, and developers will quickly test whether Hy4 matches the company’s claims outside its own benchmarks. Second, the cheaper Hy4 Flash, which will show how aggressively Tencent wants to compete on price. Third, the safety debate: independent researchers will probe the model for offensive cybersecurity and biological risks, and their findings — not the launch announcement — will decide whether Hy4’s defensive-training approach becomes a template other open-model makers follow.
Sources: Tencent announcement via PRNewswire / The Korea Herald, 4 November 2026; Reuters, 4 November 2026. Published: 4 November 2026, Kuala Lumpur time.
