AI
Tencent releases open-source large model Hy4 and introduces vision and computer-operation capabilities
Tencent released its next-generation large language model Tencent Hy4, saying it ranks among the world’s leading open-source models, and will release model weights and a system card in the coming days. Hy4 gains visual understanding and computer-operation capabilities for the first time, strengthening long context, agentic tasks, and scenarios including office work and finance.
Tencent announced the release of its next-generation large language model Tencent Hy4, saying its performance and capabilities have improved significantly and that it has joined the ranks of the world’s leading open-source AI models. Tencent Hy said it will release model weights and a system card in the coming days, with a focus on cybersecurity and biosecurity, to reflect responsible open-source development.
Hy4 introduces visual understanding for the first time and improves instruction following, long-context understanding, and long-horizon agentic tasks. With visual understanding, reasoning, and tool calling, it can handle tasks such as fine-grained image processing, complex chart analysis, and mathematical visual reasoning.
The model also gains computer-use capability for the first time and can interact with command-line interfaces and graphical user interfaces to complete work and everyday tasks. Hy4 has been specially trained for professional software such as Blender and CAD to strengthen its ability to plan and execute complex long-horizon tasks.
In Tencent’s WorkBuddy scenario, Hy4 is used for visual creativity, 3D scene construction, web development, and human-machine collaboration. Relevant evaluations had an average score of 83.18 and a pass rate of 96.15%, with Tencent saying it ranked first in 9 of 12 scenarios in the Chinese mainland market. The model also connects to WeCom through WorkBuddy, aggregating workflow information such as chats, emails, documents, calendars, and meetings to track projects, to-dos, meeting preparation, and deadlines.
For financial analysis, Tencent said Hy4 was trained on a large number of professional workspaces and real-world scenario tasks, covering fund research, securities analysis, and quantitative investment strategies. A blind test conducted by the Tencent Hy team in collaboration with Tencent Financial Technology involved 213 financial professionals, including 153 external institutional experts. Available materials say Hy4 performed as the best open-source model in the test, but did not provide complete evaluation details.