Tencent launches Hy4 with vision and computer-use capabilities

What happened

On November 4, Tencent launched Hy4. The update gives its flagship AI model the ability to look at images and to operate computer tools. The model can read charts and documents, compare several images at once, and use command-line tools (text-based controls for computers) and graphical interfaces (the buttons and windows you click on screen).

Key numbers and facts

• The model has 770 billion parameters (the internal settings that help an AI learn and make decisions).
• Standard API price: $0.834 per million input tokens and $2.501 per million output tokens.
• A 50% discount applies on Tencent Cloud TokenHub until December 31.
• Free usage quota in WorkBuddy and CodeBuddy runs from November 4 to November 15.
• Tencent plans to release the model weights (the saved learned settings) and a system card (a document explaining how the model works and its limits) soon.

Some background

Hy4 is an AI model, which is a computer program trained on huge amounts of data to answer questions and complete tasks. Tencent trained Hy4 for specific jobs, such as working with Blender (a 3D design tool) and CAD software (programs engineers use to design objects). It was also trained on tasks that use several programs at once. The company plans a cheaper version called Hy4 flash.

Why it matters

This shows that AI models are moving beyond chat. They can now look at pictures and use software on their own, which can save people time on office work, money analysis and game making. Because Tencent plans to share the weights, developers outside the company may be able to study and build on the model.

What to watch

• Whether the planned release of the model weights and system card happens soon.
• When the cheaper Hy4 flash model comes out and how much it costs.
• How well Hy4 performs on real office, finance and game development work.
• Whether the discounts and free quotas lead to wider use after they end.