Skip to main content

Vision inputs

Vision-capable models accept images inside the content array. The gateway passes image parts through in the surface's native shape.

Image by URL​

{
  "model": "gpt-5.1",
  "messages": [
    {
      "role": "user",
      "content": [
        {"type": "text", "text": "What's in this diagram?"},
        {"type": "image_url", "image_url": {"url": "https://example.com/arch.png"}}
      ]
    }
  ]
}

Image by base64​

Inline data URLs work when the image isn't publicly reachable:

{"type": "image_url", "image_url": {"url": "data:image/png;base64,iVBORw0…"}}

Keep payloads reasonable — requests over the gateway's body cap are rejected with 400 before reaching any provider. For repeated use of the same asset, host it and pass the URL.

Practical notes​

  • Filter the model catalog by the Vision capability tag.
  • Images count toward input tokens at provider-specific rates; the usage block on the response reflects the true billed amount.
  • The console Playground supports image attachments end-to-end — useful to sanity-check a model's vision behaviour before writing code.