Tencent said on the 5th that it will expand the global availability of its artificial intelligence (AI) model Hy3 (formerly Hunyuan), which it released on the 6th on the. As a result, developers can now connect Hy3 to development environments, coding extensions, and external platforms through an application programming interface (API).
General users and corporations can access Hy3 through Tencent's AI work tool WorkBuddy, and it is available free to users worldwide through the 31st. Hy3 is also available through Tencent Design Miura and Tencent Cloud TokenHub, and integration will expand across Tencent Cloud's AI-native product portfolio going forward.
Hy3 is a model that applies a mixture-of-experts (MoE) architecture. It has a total of 295 billion parameters and activates 21 billion parameters per task. It supports a context length of up to 256K.
Tencent said Hy3 recorded more than 68 times as many API calls as the previous-generation model within a week of its release, and ranked No. 1 worldwide in global LLM usage on OpenRouter during the same period. Hy3 is currently available on external developer platforms such as Hermes and Kilo, Klein, OpenClaw, OpenCode, and Cherry Studio, and the model has also been published on Hugging Face and ModelScope.
In evaluations conducted in Tencent's internal work environment, applying Hy3 to WorkBuddy pushed the task success rate above 90%, while the average task completion time was reduced by 34% compared with the previous-generation Hy model.
Miura, newly integrated with Hy3, is an AI-native creative studio for design teams, content creators, and marketers. By integrating Miura with Hy3, Hy3's reasoning and agent capabilities can be applied to complex creative workflows while keeping the design process in an editable and traceable form.
On OpenRouter, Hy3 API pricing starts at $0.1288 per 1 million input tokens and $0.5336 per 1 million output tokens.
Poshu Young, senior vice president of Tencent Cloud and head of Tencent Cloud International, said, "Hy3 represents a significant step forward in helping put AI to work in real-world tasks in a practical, reliable, and scalable way," and added, "By connecting advanced models to real-world deployment environments, enterprise-grade engineering, flexible model orchestration, and secure infrastructure, we aim to provide corporations with the control and stability needed to move from pilot projects to large-scale deployment."