
Tencent has released Hy4, the full version of the AI model it previewed in August, adding visual understanding and computer-use capabilities as it pushes deeper into workplace software. In its 4 November announcement, the Chinese technology group said the model is available through Tencent Cloud TokenHub and several of its own products. Model weights and system cards are promised in the coming days, but Tencent has not yet said they are available today.
That distinction is important for developers deciding whether Hy4 can be deployed independently. Tencent calls the release part of its open-source strategy, while the immediate route to the new model is through hosted services and applications. The company says it will publish more details on safety and release the weights shortly.
What Hy4 adds
Hy4 can examine images, charts and other visual material, according to Tencent. It also adds the ability to operate command-line and graphical interfaces, potentially allowing an AI agent to carry out longer sequences of actions in software rather than merely answer questions. Tencent says it trained the model on professional workflows involving tools such as Blender and CAD. These are company claims, and real-world reliability will depend on the task, the access granted to the agent and the safeguards around it.
The release builds on Hy4 Preview, which Tencent open-sourced in August with 770 billion total parameters and 49 billion active parameters. The preview already targeted coding, office work and scientific research. The full release shifts the pitch further towards agents that can read visual information and act inside workplace applications.
Tencent cites internal evaluations in finance, game development and office productivity. In one company-run comparison of 579 financial tasks reviewed by professionals, Hy4 scored ahead of several rival open-source models. Such results give a sense of where Tencent wants to compete, but they are not a substitute for independent testing across different languages, organisations and safety settings. Tencent also reports a lower hallucination rate than the preview in its own evaluation.
Availability and pricing
Hy4 is available through Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom’s AI Assistant and other Tencent services. Tencent says third-party API availability is being added through platforms including OpenRouter. Its stated standard price is approximately $0.90 per million input tokens and $2.70 per million output tokens, with cached input priced at about $0.045 per million tokens. A 50% TokenHub promotion runs through 31 December, while WorkBuddy and CodeBuddy offer free access from 4 to 15 November. Tencent says a more economical Hy4 Flash model is due soon, without specifying a date.
For businesses, the larger question is whether the model can perform useful work reliably across documents, applications and teams. Computer use raises the potential value of an AI assistant, but it also raises the stakes when an agent has permission to read private material or change a system. Tencent says its safety work has focused on cybersecurity and biological risk; developers will still need to test permissions, logging and human approval in their own deployments.







