Vercel Adds Hy4 Preview to AI Gateway for Long-Context Workloads.

Vercel has added Tencent’s Hy4 Preview to AI Gateway, giving developers access to a large open-source model designed for long-running coding, document analysis and scientific reasoning tasks. The model combines a 1 million-token context window with a Mixture-of-Experts architecture, making it a potentially useful option for applications that need to process unusually large codebases or collections of documents.

The company announced the integration on August 28, 2026. Hy4 Preview can now be called through Vercel’s AI SDK, selected inside supported coding agents, or tested through the gateway’s model playground.

Hy4 Preview has 770 billion total parameters, but activates 49 billion parameters for each token. That Mixture-of-Experts design routes individual requests through a subset of the model rather than engaging every parameter for every token. Vercel says the model is aimed at long-horizon coding, game development, document analysis and scientific reasoning, areas where preserving context across lengthy inputs can matter.

The 1 million-token context window is the most immediately relevant specification for builders. It creates room for large repositories, extensive technical documentation, lengthy research material or combinations of code and supporting files within a single context. A large context window does not by itself establish output quality or reliability, particularly while the model remains in preview, but it gives developers another model to evaluate for workloads that exceed conventional context limits.

Potential uses include:

• Reviewing or modifying code across a large project
• Analysing long contracts, reports or document collections
• Maintaining context through multi-step coding tasks
• Working with detailed game specifications and related source files
• Supporting research workflows that combine multiple technical sources

Developers using the AI SDK can access the model by setting the model identifier to tencent/hy4-preview. Vercel’s example uses streamText:

import { streamText } from 'ai';

const result = streamText({
  model: 'tencent/hy4-preview',
  prompt: 'Add pagination to the results endpoint.',
});

That relatively small configuration change should make Hy4 easier to test alongside other models already exposed through AI Gateway. Teams should still evaluate response quality, latency and suitability using their own prompts before relying on a preview model for production work.

Hy4 Preview can also be connected to coding agents using Vercel’s coding-agent setup command. The company lists Claude Code, Codex, OpenCode, Cursor, Pi and other compatible agents. After running vercel ai-gateway coding-agents setup, a developer can select tencent/hy4-preview from inside the agent.

The release is part of Vercel’s broader effort to turn AI Gateway into a common access layer for models with different strengths. Recent additions have included Muse image generation and editing and Gemini 3.5 Transcribe. Vercel has also added per-user AI Gateway budgets, providing teams with a way to cap individual spending while experimenting with this expanding model catalogue.

For small teams, the practical benefit is choice without building a separate integration for every model. Hy4 Preview broadens that choice for long-context and agentic coding scenarios. Its preview label remains important, however. The announcement does not specify general-availability timing or model-specific pricing, so teams considering sustained production use will need to watch for further details.

What is Vercel Hy4 Preview?

Hy4 Preview is Tencent’s open-source Mixture-of-Experts model, now available for inference through Vercel AI Gateway. It has 770 billion total parameters, activates 49 billion per token and supports a 1 million-token context window.

How can developers use Hy4 Preview?

Developers can set model to tencent/hy4-preview in Vercel’s AI SDK. They can also choose the model in Vercel’s model playground or connect it to supported coding agents after running vercel ai-gateway coding-agents setup.

Is Hy4 Preview ready for production use, and what does it cost?

Vercel describes the release as a preview and has not provided a date for general availability in the announcement. The changelog entry also does not specify model pricing, so developers should consult the current AI Gateway model page and test the model against their own requirements before production adoption.