Building on Hy3 preview, the official Hy3 delivers enhanced model performance, greater stability and cost efficiency, and expanded integration across Tencent products and services

Tencent today officially launched Hy3. The model demonstrates significantly enhanced performance relative to models of the same size, while achieving intelligence comparable to flagship models with two to five times its parameter scale. Compared with the preview version, it also delivers greater stability and cost efficiency. 

Hy3 has already been adopted across Tencent products and services, including WorkBuddy/CodeBuddy, Yuanbao, Marvis, ima and others. Its API is now available on Tencent Cloud TokenHub, with global third-party developer platforms set to integrate Hy3 progressively.

Hy3 is a hybrid fast-and-slow-thinking model built on a Mixture-of-Experts (MoE) architecture, with 295 billion total parameters and 21 billion active parameters, supporting a context length of up to 256K. Its preview version, launched on 23 April as the first model in the Hy3 generation, made significant improvements over Hy2 in complex reasoning, instruction-following, in-context learning, code generation and agent capabilities.

Building on the foundation of Hy3 preview, Hy3 takes another leap across a wide range of tasks through strengthened reinforcement learning and enhanced data quality and diversity, delivering performance comparable to larger-scale flagship models. Hy3 has made particularly notable progress in productivity tasks such as software development, office productivity, financial modeling, front-end design and game production, making it a reliable and cost-efficient choice.

Refined through feedback from real-world use to deliver an enhanced user experience 

Hy3 has been refined through use by global developers and across Tencent's large-scale real-world business scenarios. Since the launch of Hy3 preview, its average daily token consumption has increased twenty-fold, reflecting growing market recognition of its positioning as a practical and cost-effective model.

As one of the most widely used AI productivity agents in China, WorkBuddy is built around practical and complex workplace scenarios, including automated script generation and workflow orchestration. These real-world use cases provide valuable input for the continued evolution of Tencent Hunyuan (now known as Tencent Hy) models. Since launch, the number of WorkBuddy users who actively select Hy3 preview has grown sixfold.

Building on Hy3's significantly enhanced Agent capabilities, Yuanbao has also introduced an Agent function. Users can simply describe what they need in natural conversation, and Yuanbao can help perform complex tasks and generate files in formats such as PowerPoint, Word, Excel, PDF and HTML, supporting everyday workplace needs at no cost.

Hy3 has also strengthened ima’s knowledge-base Q&A to deliver more structured reasoning, broader information coverage, and more complete, structured and usable outputs for long-form writing. In Agent scenarios, Hy3 improves its task planning and tool coordination.

For Marvis Agent, Tencent's OS-level AI assistant, Hy3 enhances core capabilities such as file editing and generation, file management, computer diagnostics and task execution. It also supports more reliable multi-agent collaboration, improving speed, stability and cost.

Hy3 also supports a range of business scenarios across Weixin/WeChat and games. On Weixin Official Accounts, AI assistants for customer service powered by Hy3 can better understand user intent, enabling more accurate responses to vague or incomplete questions by taking earlier context into account, instead of relying on templated replies. 

In gaming, Path of Exile: Advent on WeGame has also connected its AI assistant to Hy3, helping it respond to player requests more accurately and reliably while reducing hallucinations and improving the overall player experience. 

Making AI more accessible to users and enterprises through a practical and open approach

On the open-source front, Hy3 is available under the commercially friendly Apache 2.0 license, allowing developers worldwide to download and use it. To make Hy3 more accessible to global developers, the model will be progressively made available on global third-party developer platforms, including OpenRouter, Hermes, Kilo, Cline, OpenClaw, OpenCode and Cherry Studio. It has also been available from day one on open-source model communities including Hugging Face and ModelScope.

From rebuilding its infrastructure in late January 2026, to the launch of Hy3 preview in April, and now to the latest progress of Hy3, Tencent Hy has completed its full model-development loop in less than six months, spanning foundational reconstruction, product feedback and model improvement. Looking ahead, Tencent Hy will continue to accelerate its technical iteration, push the boundaries of model intelligence, and focus on real-world applications, turning leading large model capabilities into accessible productivity for every industry.