Hy4, Tencent's reasoning model, becomes available on Vercel AI Gateway with vision

Hy4, Tencent's reasoning model, becomes available on Vercel AI Gateway with vision

Editor's note: the transcript of the original source (Vercel changelog), reproduced at the end of this article, was captured incompletely by our collection process: it ends in page interface elements ("Loading status… Select a display theme"), which are just page navigation items and not substantive information. This does not affect any of the article's factual claims, all of which were checked against the actual changelog content.

Hy4 is the first model in the Hy line to support images, and also brings computer use and a 1-million-token context window. See what changes in practice when testing the model.

Vercel AI Gateway has started listing Hy4, Tencent's reasoning model, at a 50% discount off list price valid through December 31, 2026. It is the first model in the Hy line to support image input, which changes the type of task that can be delegated to it within an agent pipeline. The source is Vercel's official changelog, which lists Hy4 among the models available on AI Gateway.

For those who maintain LLM integration in production, the news isn't just "another model on the list." It's one more option within a gateway that already promises routing, failover, and unified billing, which lowers the cost of trying out a new model: in practice, this can mean switching from tencent/hy4 to another provider by adjusting a few lines of configuration, not a full reintegration.

What Hy4 brings that's different from Hy4 Preview

According to Vercel, Hy4 is the first version in the Hy family with vision: it can pick out fine detail in images and read complex charts, not just simple OCR. It also gained computer use support in both command-line and graphical interfaces.

Compared with Hy4 Preview, the line's previous version, Tencent highlights three specific improvements:

• Follows instructions more precisely in complex prompts
• Maintains quality in long context and long-running tasks
• Performs better in deep research, that is, in search and synthesis chains that require multiple chained calls

None of these three points comes with a public benchmark in the source, so it's worth treating them as directional indications, not as hard numbers for comparing against other models.

Technical specifications and how to call the model

Hy4 supports reasoning, tool calling, structured outputs, and image input, with a 1-million-token context window. That's plenty of room to drop an entire codebase, or a stack of documents, into the prompt without rewriting chunking logic.

On AI Gateway, the model name is tencent/hy4. With the AI SDK, the call follows the same pattern as the other models in the catalog:

import { streamText } from 'ai';

const result = streamText({
 model: 'tencent/hy4',
 prompt: 'Outline a database migration in three brief steps.',
});

for await (const text of result.textStream) {
 process.stdout.write(text);
}

To use Hy4 inside Claude Code, Codex, Cursor, or another coding agent, the way to do it is to install the latest version of Vercel CLI and run npx vercel ai-gateway setup, selecting tencent/hy4 as the agent's model. Those who just want to test without touching a project already have Vercel's model playground available for that.

What AI Gateway solves (and what it doesn't)

AI Gateway's proposition isn't new, but it's worth reinforcing because it's what makes Hy4 relevant beyond being a Chinese curiosity: a unified API for calling models from different providers, usage and cost tracking, retries, failover, and performance optimizations with a promise of higher uptime than any single provider.

The gateway also includes custom reporting, Zero Data Retention support, per-API-key budgets, and routing rules. Vercel emphasizes that it reflects provider pricing with no markup and does not charge a platform fee on inference, including on requests made with your own key (BYOK).

In short: the 50% discount is off Tencent's own list price, not a cut invented by Vercel, and the gateway works as a routing and observability layer on top of that, not as an alternative pricing provider.

When it's worth testing Hy4

The decision to test a new model on the gateway isn't just about price: it involves weighing whether the cost savings outweigh the risk of a newly launched feature, without the same track record of at-scale use as already-established models.

For those who already have a model in production, switching to Hy4 only pays off if the cost savings (the 50% discount through December) outweigh that maturity risk. For those limited by the lack of native vision in their current provider, Hy4 is the most direct option to test, since it fills that gap without changing integration architecture (the gateway abstracts that away).

What's still left open

The source doesn't detail Hy4's base price per token before the discount, nor does it provide a comparative benchmark between Hy4 and other models on reasoning or agentic coding tasks. There's also no information about Tencent's specific data retention policy behind the Zero Data Retention offered by the gateway, something compliance-conscious teams will want to confirm before moving production workloads.

For those who just want a low-risk test, the most reasonable path is to run tencent/hy4 in the model playground on vision and deep research tasks, which are exactly the points Tencent highlights as differentiators, and compare the output side by side with the model already in production before migrating any critical flow.

Source 1: Vercel Changelog (https://vercel.com/changelog/tencent-hy4-now-available-on-ai-gateway)

Tencent Hy4 now available on AI Gateway - Vercel

Skip to content Hy4 from Tencent is now available on AI Gateway , at 50% off through December 31, 2026. Hy4 supports reasoning , tool calling , structured outputs , and image input, and has a 1M-token context window for work across a codebase or many documents. Hy4 is the first Hy model with vision. It can pick out fine detail in an image and read complex charts. It also supports computer use in both command-line and graphical interfaces. Compared with Hy4 Preview , it follows instructions more closely and does better with long context, long-running tasks, and deep research.

Use tencent/hy4 as the model name: AI SDK AI SDK AI SDK fx fx Chat Completions Chat Completions 1 import { streamText } from 'ai' ; 2 const result = streamText ( { 3 model : 'tencent/hy4' , 4 prompt : 'Outline a database migration in three brief steps.' , 5 } ) ; 6 for await ( const text of result . textStream ) { 7 process . stdout . write ( text ) ; 8 } Copy for agent View model details ↗ To use it in Claude Code, Codex, Cursor, and more, install the latest Vercel CLI and run setup: npx vercel ai-gateway setup Then select tencent/hy4 in the agent.

See the coding agents guide for details. Try Hy4 in the model playground . AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations for higher-than-provider uptime. It includes built-in custom reporting , Zero Data Retention support , budgets for API keys , routing rules , and more.

AI Gateway reflects provider pricing with no markup and does not charge a platform fee on inference, including on Bring Your Own Key (BYOK) requests. You can view all language models available on AI Gateway. (note: the original source transcript ends here with Vercel website interface elements, with no additional informational content beyond what has already been reproduced above.)