Originally published on The Daily Flare.

Tencent, the Shenzhen-based technology company, launched Hy4 on Nov. 4, 2026, adding visual understanding and computer-use features to its flagship language model.

The company says the new model can interpret images, analyse complex charts and perform mathematical reasoning with visual information. It can also operate command-line tools and graphical interfaces, extending Hy4 beyond text-based tasks to work that involves interacting with software.

Tencent said Hy4 was trained to use professional applications including Blender and computer-aided design software. The model is intended to plan and carry out complex tasks that can extend across multiple steps. TechNode described it as a 770-billion-parameter model, while Tencent said it has a context window of one million tokens, the same length as Hy4 Preview.

The release builds on Hy4 Preview, which Tencent introduced on Aug. 28. SiliconANGLE reported that the preview had 770 billion parameters, with 49 billion active parameters. Tencent said it expanded both pre-training and post-training for the new model, and reported improvements in instruction following, long-context comprehension, extended agent tasks and deep research. It also identified excessive reasoning and over-verification in the preview’s thinking process as areas addressed in the new version.

Workplace and software tasks

Tencent is positioning Hy4 for workplace productivity, collaboration and professional work. The company said its long-context capabilities were optimized to handle complex information and tasks such as web development, document summarization, visual design and presentation creation. It also cited improvements in front-end and game development, including work involving business logic, state management and coordination among components.

Tencent said Hy4 was integrated with Unreal Engine, Unity and Godot to support complex game logic and multiplayer development. Compared with the preview, it reported a 19.1 percent improvement on its GameDev benchmark and a 9.8 percent gain on GameLogic. The company did not publish the underlying baseline scores, SiliconANGLE reported.

Hy4 is also integrated into WorkBuddy, Tencent’s workplace AI tool. Tencent reported an average score of 83.18 and a 96.15 percent pass rate on WorkBuddy evaluations. It said the model ranked first in nine of 12 scenarios in the mainland Chinese market, and described improvements in document presentations, stability and usability during human-AI collaboration.

When connected with WeCom through WorkBuddy, Hy4 can work with information from chats, email, documents, calendars and meetings, Tencent said. The company described uses including keeping track of project history, identifying outstanding tasks, flagging meetings that need preparation and monitoring upcoming deadlines. Tencent said team collaboration features using Hy4 are planned for WorkBuddy’s upcoming Projects feature.

The company also presented results from a financial-analysis evaluation conducted by the Tencent Hy team with Tencent Financial Technology. The blind assessment covered 579 tasks based on real-world financial work and involved 213 professionals, including 153 experts from outside Tencent. About 70 percent of participants had more than five years of industry experience. Tencent said Hy4 scored 3.04, ahead of DeepSeek-V4.1-Flash-1017 at 2.97, Kimi K3.1 Pro at 2.68 and GLM-5.3 at 2.68.

Availability and planned releases

Tencent said Hy4 is available through Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao, WeCom’s AI Assistant and Ima. It can also be accessed through third-party platforms including OpenRouter, Hermes and OpenCode.

The company said it plans to release the model weights and system cards in the coming days, describing the release as part of its open-source AI development effort and citing cybersecurity and biological safety as areas of focus. Tencent also said a lower-cost Hy4 Flash version is scheduled to follow.

Despite the new features, Tencent’s launch materials frame Hy4 as an open-weight model rather than a match for every leading closed-source system. The South China Morning Post reported that Tencent acknowledged Hy4 still lagged the world’s leading closed-source models on challenging internal tests involving research-level mathematics, long-horizon coding, model self-improvement and advanced visual question answering.