Images

Current status

The API accepts text content only. A message's content can be a plain string, or an array of {"type": "text", "text": "..."} parts. A message that includes any other part type, including an image_url part, is rejected before any generation happens, and this applies to every model tier.

What you will see

Sending an image part returns a 400 in the standard error envelope described on the Errors page:

HTTP/1.1 400 Bad Request

{"error": {"message": "messages must be a non-empty array of {role, content} objects. ...", "type": "invalid_request_error"}}

In the meantime

If your application needs the model to reason about an image, do the extraction step on your side (OCR, captioning, or a description written by the user) and send the result as ordinary text in the message. This page will be updated if image input becomes available.