Tencent launches Hy4, an open-source AI model with computer vision and graphical interaction capabilities

1–2 minutes

On Wednesday, 4th, Tencent unveiled Hy4, an open-source artificial intelligence system that the company claims positions itself among the leading models in the market. The launch highlights new capabilities in visual understanding, computer interaction, and advancements in professional tasks.

New features

Hy4 has been trained to understand images, interact with graphical interfaces, and execute commands via text input. According to the company, the model received specialized training in programs like Blender and CAD tools, enhancing its applicability in design and engineering scenarios.

Performance and comparisons

In internal tests, Hy4 outperformed competitors in financial tasks and in scenarios involving WorkBuddy, a Tencent-developed assistant. The company emphasizes that these results are preliminary and do not constitute independent validation of the model’s superiority in any specific task.

Technical background

Hy4 evolved from a preview version released in August. At that time, Tencent detailed an architecture with 770 billion total parameters, of which 49 billion are activated during processing, and a context window exceeding 1 million tokens. This capability enables handling larger volumes of information in a single interaction.

The model was designed to support programming, office, and research tasks and serves as the foundation for products like CodeBuddy and WorkBuddy, which have collectively received over 9 billion task requests since March, with adoption across more than 50 sectors.

Corporate strategy

The launch is part of Tencent’s broader strategy to provide AI solutions to businesses. The company also announced TokenHub, a unified platform offering access to its models and third-party alternatives through a single API, giving organizations the flexibility to choose between performance and cost.

Tencent stated that the model’s weights and technical security documents will be released in the coming days, fulfilling Hy4’s promise of openness.

Frequently Asked Questions

What has changed in Hy4?

Tencent highlights improvements in image understanding, computer interaction capabilities, and performance in long-duration tasks.

Does a good test result guarantee performance in any use case?

No. It is essential to consider which tasks were evaluated, how the test was conducted, and whether the results are reproducible in real-world use.