Tencent has released Hy4 preview, a new open-source AI model with 770 billion parameters and a context window exceeding 1 million tokens. The model is designed less as a general chatbot novelty and more as a productivity engine for coding, office work, scientific research and data-heavy tasks.
The release adds another major contender to China’s rapidly evolving open-model ecosystem—and shows how quickly Tencent is accelerating its AI development cycle.
What Is Tencent Hy4 Preview?
Hy4 preview is the latest large language model from Tencent’s Hunyuan AI team. It uses a Mixture-of-Experts (MoE) architecture with 770 billion total parameters, but only about 49 billion parameters are activated for each token.
That distinction matters.
Instead of running the entire 770B-parameter model for every request, the system routes each token through selected “experts.” This can provide the capabilities of a very large model while reducing the computation required for individual queries.
Tencent says Hy4 preview also supports a context length of more than 1 million tokens, allowing it to process exceptionally large amounts of information within a single interaction.
What Can Hy4 Preview Do?
Tencent is positioning the model around real-world productivity, rather than simply conversational ability.
Its target applications include:
- Software engineering and coding
- Office and workplace tasks
- Scientific research
- Data analysis
- Game development
- Financial and security-related workloads
Tencent has also made Hy4 preview available through products including WorkBuddy and CodeBuddy, while developers can access the model through Tencent Cloud’s TokenHub and OpenRouter.
The model is released under the Apache 2.0 licence, according to its Hugging Face repository, making it available for broader developer experimentation and deployment.
Why 770 Billion Parameters Matter
The headline number—770B parameters—is impressive, but it should not be interpreted as a direct measure of intelligence.
The more interesting feature is the combination of 770B total parameters + 49B active parameters + 1M-token context.
Hy4 preview uses 78 layers, with most built around MoE architecture. Its design also incorporates techniques including Gated DeepSeek Sparse Attention and IndexCache to make attention processing more efficient.
In practical terms, Tencent is trying to build a model that can handle large, complicated tasks without activating its entire neural network for every token.
Why Tencent Released Hy4 So Quickly
The timing is notable.
Tencent released Hy3 in July 2026, making Hy4 preview a remarkably rapid next-generation update. Chinese AI companies are increasingly competing not only on model quality but also on how quickly they can release improved open models.
The release also comes as Tencent increases its focus on AI products and infrastructure, placing Hunyuan in direct competition with other Chinese model developers as well as leading US AI companies.
The Bigger AI Race
Hy4 preview highlights a broader shift in the AI industry: open-source and open-weight models are becoming increasingly capable at specialised professional tasks.
The important question is no longer simply which company has the biggest model.
It is increasingly about who can build models that developers can access, customise and deploy cheaply—and make them useful for real work.
With Hy4 preview, Tencent is clearly betting that the next stage of China’s AI race will be fought not just in chatbots, but in codebases, laboratories, offices and autonomous AI workflows.