Invalid type for 'input' when sending array input to /v1/responses

Nobody has claimed this yet.

When sending a request to the /v1/responses endpoint with input as an array (to include multiple content types such as text and image), the API returns an error stating that the input type is invalid.

This behavior makes it impossible to send multimodal (text + image) queries via the responses endpoint, even though the structure mirrors how /v1/chat/completions handles messages.

Request
`{
"model": "gemma-3-12b-it",
"input": [
{
"role": "user",
"content": [
{
"type": "text",
"text": "What is in this image?"
},
{
"type": "image_url",
"image_url": "{{TestBase64Image}}"
}
]
}
]
}
`

Response:
`{
"error": {
"message": "Invalid type for 'input'.",
"type": "invalid_request_error",
"param": "input",
"code": "invalid_union"
}
}
`

Review the /v1/responses documentation and its input schema, then compare it with the /v1/chat/completions message format shown in the report. Reproduce the request using the provided multimodal example and determine whether the documented input shape accepts arrays. Done means the endpoint behavior and supported input format are accurately documented or corrected.

Written by the indexing model from the issue text.