When to use a vision coding agent instead of a plain LLM?
Developers wiring coding agents into a large codebase often hit two walls: models that cannot read images or screenshots, and models that lose the thread on long-running tasks. Vercel's AI Gateway now offers Tencent Hy4, a model built for exactly that mix, with vision, computer use, and a 1M-token context window.
What it is
Hy4 is Tencent's first Hy model with vision, available on AI Gateway under the model name tencent/hy4. It supports reasoning, tool calling, structured outputs, image input, and computer use in both command-line and graphical interfaces, at 50% off through December 31, 2026.
Applied case
A developer using Claude Code, Codex, or Cursor can install the latest Vercel CLI, run setup, and select tencent/hy4 in the agent. Because it carries a 1M-token context window, it can work across an entire codebase or many documents at once, and per Vercel it follows instructions more closely than Hy4 Preview, especially on long context, long-running tasks, and deep research. Its vision lets it pick fine detail out of an image and read complex charts, useful when an agent must interpret UI screenshots or architecture diagrams.
- Use tencent/hy4 in Claude Code, Codex, or Cursor after running Vercel CLI setup; 1M-token context covers a whole
When to use it
Use when your agent needs vision, computer use, or sustained work over a huge context.
Skip when a cheaper text-only model already handles your task.
Read more
Recommended for you
- WorkflowVercel
How I get usage credit without juggling a dedicated Token Pass key
Token Pass credit now applies automatically to AI Gateway requests, no special key required, with fallback to team credits when it runs out.
- WorkflowOpencode
How do you keep coding agents on task without babysitting every step?
Agent sessions drift from the objective, forcing you to watch every turn; opencode's /goal lets a session run until a reviewer confirms the goal is met.
- WorkflowHerdr
How do you supervise coding agents without watching every window?
I run Claude and opencode side by side in one terminal. Three herdr habits replaced window-polling with state-waiting: persistent sessions, the agent loop, a setup script.
- WorkflowSuperpowersClaude Code
Why does your multi-agent workflow keep colliding?
Two agents in two threads share files but not context. Both decide on stale state. Fix: one fresh agent per task, with isolated context.
Enjoyed this article?
Subscribe for new articles. No spam. Unsubscribe anytime.