Tencent Hy4 Preview Launches as a 770B Open-Source AI Model With 1M+ Context

Tencent Opens Hy4 to Developers With Massive Context Window and Strong Coding Capabilities

The launch comes only weeks after Tencent expanded global access to its previous Hy3 model.

That makes Hy4 particularly interesting because Tencent is not simply adding another experimental model to its lineup. The company is attempting to position Hy4 as a practical open-source system for developers, businesses and users who need advanced reasoning, coding and long-context capabilities.

What Is Tencent Hy4 Preview?

Tencent Hy4 Preview is the latest model in Tencent's Hy AI family.

The company says Hy4 is built around real-world productivity rather than focusing only on benchmark performance. Its capabilities are aimed at areas such as software development, office work and scientific research.

The model has:

• 770 billion total parameters
• 49 billion active parameters
• More than 1 million tokens of context
• Open-source availability
• Coding capabilities
• Office productivity capabilities
• Scientific research capabilities
• Access through multiple Tencent products
• API access through Tencent Cloud TokenHub and OpenRouter

The large context window is one of Hy4's most notable technical characteristics.

A context window exceeding 1 million tokens allows an AI model to process extremely large amounts of information within a single task, although practical performance will still depend on the application, input format and workload.

Why the 1M+ Token Context Matters

Long-context AI models are becoming increasingly important as users move beyond simple questions and short documents.

A developer may want an AI system to understand a large software repository.

A researcher may need to analyze hundreds of papers.

A business user could work with extensive reports, contracts or internal documentation.

Traditional context limits can force users to divide these workloads into smaller pieces.

Hy4's context window exceeding 1 million tokens is designed to reduce that problem.

Tencent specifically highlights the model's usefulness for real-world productivity tasks, including coding, office work and scientific research.

However, a large context window should not automatically be interpreted as meaning the model will perfectly understand every million-token input.

Long-context capability is an important technical feature, but model quality, retrieval accuracy, reasoning and application design remain equally important.

Hy4 Has 770 Billion Parameters

Hy4 Preview contains 770 billion total parameters, with 49 billion active parameters, according to Tencent.

The difference between total and active parameters is important.

Modern large AI models can use architectures where only part of the model is activated for a particular request. This can allow extremely large models to provide substantial capacity without requiring every parameter to participate in every token calculation.

Tencent has not positioned Hy4 simply as a model with a huge parameter count.

Instead, its announcement emphasizes practical performance across coding, office tasks and scientific research.

That focus reflects a broader change in AI development.

Model makers are increasingly competing not just on size, but on how effectively models perform useful work.

Hy4 Is Open Source

One of the biggest reasons developers may pay attention to Hy4 is its open-source availability.

Tencent says Hy4 Preview has been released and open-sourced, allowing developers to work with the model rather than accessing it only through a closed consumer chatbot.

Open models can be particularly valuable for:

• AI developers
• Researchers
• Enterprise teams
• AI startups
• Coding-tool developers
• AI infrastructure companies
• Organizations that want more control over deployment

Open-source availability also makes it easier for the wider AI community to evaluate the model, build integrations and experiment with different use cases.

That makes Hy4 part of the larger competition between proprietary frontier models and increasingly capable open AI systems.

Hy4 Is Available Through Tencent Products

Tencent is not limiting Hy4 to a developer repository.

The company says users can access Hy4 globally through products including WorkBuddy and CodeBuddy, as well as Yuanbao and ima. Developers can also connect to the model through Tencent Cloud TokenHub and OpenRouter.

This gives Hy4 several potential entry points.

A regular user may interact with it through a Tencent application.

A developer can integrate it through an API.

A researcher can experiment with the open-source model.

A company could potentially build a customized workflow around the model.

This combination of open-source access and commercial product integration is an important part of Tencent's strategy.

Hy4 Targets AI Coding

Software development is one of the most important areas Tencent is targeting with Hy4.

Tencent specifically lists coding among the model's core real-world productivity capabilities.

A large context window can be especially useful for coding because modern applications may contain thousands of files and large amounts of documentation.

An AI coding assistant needs more than the ability to generate individual functions.

It increasingly needs to understand:

• Project architecture
• Existing source code
• Documentation
• Dependencies
• Configuration files
• Tests
• APIs
• Multiple related files

Hy4's long context could make it useful for workloads where developers need an AI system to understand large amounts of project information at once.

That does not mean it will automatically outperform every specialized coding model.

Actual developer performance will depend on benchmarks, tool integration, latency, cost and real-world testing.

Scientific Research Is Another Target

Tencent also highlights scientific research as a target application for Hy4.

Research workflows can involve extremely large collections of information.

Scientists and analysts may need to compare papers, inspect datasets, summarize technical documents and reason across large amounts of context.

A model capable of processing very large inputs could potentially simplify some of these workflows.

For example, researchers could use long-context AI to analyze extensive documentation or connect information spread across multiple sources.

Again, a large context window is not a replacement for reliable scientific reasoning.

Researchers still need to verify AI-generated conclusions, especially in areas where mistakes can have serious consequences.

Hy4 Also Targets Office Work

Tencent is positioning Hy4 as a general productivity model rather than a coding-only system.

Office work is explicitly included among the model's target areas.

That could include tasks such as:

• Document analysis
• Report preparation
• Information extraction
• Research
• Business writing
• Data interpretation
• Long-document summarization

The ability to process very large inputs could be useful for organizations that work with lengthy business documents.

For example, an enterprise could potentially provide an AI system with extensive project documentation and ask it to identify relevant information without repeatedly splitting the material into smaller prompts.

Hy4 Preview Is Free on Some Tencent Products

Tencent says Hy4 Preview will be available free for two weeks on WorkBuddy and CodeBuddy after launch. Tencent also extended free access to Hy3 on those platforms until September 30.

This gives users an opportunity to test the new model without immediately paying for access.

However, users should distinguish between free access through Tencent's applications and the economics of API or self-hosted usage.

Open-source availability does not necessarily mean that running a model of this scale will be free.

Infrastructure, hardware, storage and inference costs can still be significant.

How Hy4 Compares With Tencent Hy3

Hy4 follows Tencent's Hy3 model, which Tencent released earlier in 2026.

Tencent announced global access to Hy3 on August 5, describing it as a 295-billion-parameter model with 21 billion active parameters and a 256K-token context window.

Hy4 increases the scale substantially.

This comparison shows why Hy4 represents a significant step within Tencent's model family.

The jump from a 256K context window to more than 1 million tokens is particularly notable for long-document and large-codebase workloads.

What Hy4 Means for the Open AI Model Race

Tencent's release adds more pressure to an already competitive open-model ecosystem.

Developers now have access to an expanding range of large models from companies and research organizations around the world.

The competition is increasingly moving toward practical capabilities:

• Coding
• Reasoning
• Agents
• Long-context understanding
• Multimodal workflows
• Enterprise deployment
• Lower inference costs
• Open-source availability

Hy4's combination of large scale, long context and open availability gives it a potentially interesting position.

Its real impact, however, will depend on independent evaluations and how developers perform with it in real applications.

Should Developers Try Tencent Hy4?

Developers interested in open AI models should consider testing Hy4, especially if their workloads involve large amounts of code or documentation.

The model is particularly interesting for users who need:

• Very long context
• Open-source access
• Coding assistance
• Research workflows
• Business-document analysis
• Tencent's AI ecosystem
• API-based experimentation

The best approach is to test Hy4 against the models already used in a particular workflow.

A model that looks impressive on paper may not necessarily be the best option for every application.

Developers should compare accuracy, latency, cost, context handling and tool integration before moving a production workload.

The Bigger Picture

Tencent's Hy4 release highlights how quickly the open-model landscape is developing.

Only a few years ago, models with hundreds of billions of parameters and million-token context windows were unusual.

Now they are becoming part of the competitive AI landscape.

Hy4 is especially interesting because Tencent is connecting open-source model availability with consumer applications, developer tools and cloud APIs.

That approach could help the company reach both developers and everyday users.

The next question is whether independent testing will confirm Tencent's claims about Hy4's real-world performance.

If the model performs strongly while maintaining useful inference economics, Hy4 could become an important option for developers looking for an open alternative for coding, research and long-context workloads.

For now, Tencent Hy4 Preview is one of the most significant new open-model releases to watch following its August 28 launch.

FAQs

What is Tencent Hy4 Preview?

Tencent Hy4 Preview is a next-generation open-source AI model from Tencent designed for coding, office productivity and scientific research.

How many parameters does Tencent Hy4 have?

Hy4 Preview has 770 billion total parameters and 49 billion active parameters, according to Tencent.