embed_content() raises ValueError for multiple Content objects with gemini-embedding-2 on Enterprise

[Prabindas001]

Environment details

• Programming language: Python
• OS: macOS Tahoe 26.5.1 (25F80)
• Language runtime version: Python 3.10.18
• Package version: google-genai 2.8.0

Description

When using Models.embed_content() with gemini-embedding-2 on the Enterprise Agent Platform (enterprise=True), the Python SDK raises a client-side ValueError when more than one Content object is provided in contents.

However:

• The method signature accepts ContentListUnion.
• The docstring describes contents as list[Content].
• The multimodal embeddings documentation states:

Multiple entries: Sending multiple entries in the contents array returns separate embeddings for each entry.

• The Gemini Enterprise multimodal embeddings documentation also documents multiple entries and up to 6 images per request.
• The exception is raised by the SDK before any request is sent to the backend.

Steps to reproduce

import asyncio

from google import genai
from google.genai import types

client = genai.Client(
    enterprise=True,
    project="PROJECT_ID",
    location="global",
)

MODEL = "gemini-embedding-2"

async def main():
    image_bytes = open("1.png", "rb").read()

    contents = [
        types.Content(
            parts=[
                types.Part.from_bytes(
                    data=image_bytes,
                    mime_type="image/png",
                )
            ]
        )
        for _ in range(6)
    ]

    result = await client.aio.models.embed_content(
        model=MODEL,
        contents=contents,
    )

    print(len(result.embeddings))

asyncio.run(main())

[Venkaiahbabuneelam]

Hello @Prabindas001, thanks for reaching out!

I was able to reproduce this issue. When running the exact same code through the standard Gemini API, the SDK successfully processes the list of Content objects without raising an error.

Thanks.

[Venkaiahbabuneelam]

Here I am attaching the official Multimodal Embeddings documentation,

[Prabindas001]

Thank you @Venkaiahbabuneelam for looking into this and for reproducing the behavior.

The part I'm trying to understand is whether this difference between the standard Gemini API and the Enterprise/Vertex path is intentional.

The multimodal embeddings documentation appears to state:

Multiple entries: Sending multiple entries in the contents array returns separate embeddings for each entry.

In my case, I'm passing:

contents = [
    types.Content(...),
    types.Content(...),
    ...
]

which I would expect to return multiple embeddings.

However, with enterprise=True, the SDK raises a client-side error before the request is sent:

if len(normalized_contents) > 1:
    raise ValueError(
        "The embedContent API for this model only supports one content at a time."
    )

Could you clarify whether Enterprise intentionally supports only a single Content object for gemini-embedding-2, while the standard Gemini API supports multiple Content objects, or whether the Enterprise path is also expected to support the documented batching behavior?

Thanks again for investigating.

[Prabindas001]

Just following up on this.

This clarification would be very helpful for us because we're currently evaluating the multimodal embedding architecture for a production pipeline and need to determine whether:

• multiple Content objects are expected to be supported on the Enterprise path, or
• we need to redesign around a single Content object containing multiple image parts.

At the moment, the uncertainty around the expected behavior is blocking that implementation decision.

Would appreciate any guidance on whether the current Enterprise behavior is intentional or whether support for multiple Content objects is expected.

Thanks again.

[Venkaiahbabuneelam]

Hi @Prabindas001, Thank you for the detailed explanation, and I completely understand how this uncertainty is blocking your production pipeline decisions, I need to verify with our internal engineering team whether this specific restriction on the Enterprise path is an intentional API limitation for gemini-embedding-2, or if it is a validation bug in the SDK that we plan to fix.

Thanks

[allengyrphor]

Hi @Venkaiahbabuneelam,

I'm also facing the same issue. Any updates from the engineering team on this?

Thanks.