Tencent Hy4 with vision and 1M-token context debuts on Vercel AI Gateway at half price

Tencent Hy4 arrives on Vercel AI Gateway

Vercel has added Tencent's Hy4 to AI Gateway, the company's unified routing service for language models, and is offering it at half price through December 31, 2026. The announcement, published on the Vercel blog on November 4, confirms that developers can call the model under the identifier tencent/hy4.

Vision and a million tokens of context

Hy4 is the first model in Tencent's Hy line to accept images as input. According to Vercel, it can resolve fine detail in a picture and make sense of dense charts rather than just describing a scene at a high level. Text handling gets a boost from a one-million-token context window, which Vercel frames as enough capacity to reason over a whole codebase or a large document collection in a single pass.

The model also covers the capabilities that agent frameworks depend on: step-by-step reasoning, tool calling and structured outputs. Beyond API calls, Hy4 supports computer use across both command-line and graphical interfaces, meaning it can operate software directly instead of only responding to prompts.

Vercel says that compared with the earlier Hy4 Preview, the general release follows instructions more faithfully and holds up better on long-context work, long-running tasks and deep research.

How to switch it on

Access runs through the Vercel CLI. After installing the latest version, running npx vercel ai-gateway setup lets developers select tencent/hy4 inside coding agents including Claude Code, Codex and Cursor. A model playground is also available for anyone who wants to test the model before wiring it into a project.

What the gateway provides around it

AI Gateway sits between applications and model providers, exposing one API while handling usage and cost tracking, retries, failover and performance tuning that Vercel says delivers uptime beyond what any single provider offers. The service also ships with built-in custom reporting, Zero Data Retention support, spending budgets tied to individual API keys, and routing rules.

On cost, Vercel states that the gateway mirrors provider rates without adding a margin and collects no platform fee on inference, including requests made under Bring Your Own Key.

Why it matters

Two things make this launch more than a routine catalogue addition. First, the pairing of image input with a one-million-token window is squarely aimed at agentic coding: reading screenshots and diagrams while holding an entire repository in memory is exactly the workload that tools like Cursor and Claude Code are built around, and Hy4's support for command-line and graphical computer use extends that further.

Second, the distribution channel matters as much as the model. By appearing on AI Gateway alongside Western providers, discounted by half and with no fee layered on top, Hy4 becomes cheap to evaluate for developers who would otherwise never provision a separate Tencent account. Whether it matches incumbent models on quality is something benchmarks and hands-on testing will decide, but until the end of 2026 the cost of finding out has rarely been lower.