Support `multimodalembedding@001` model

[adubovik]

Hi team,

I’d like to ask about support for the multimodalembedding@001 model in the GenAI SDK.

Currently, this model can only be accessed via the Vertex AI SDK. However, according to the documentation, this access method will no longer be available after June 24, 2026.

At the same time, the model itself is scheduled for retirement much later, on April 1, 2027.

This creates a gap where the model is still active, but there is no clear supported way to access it via the newer SDK.

Right now, calling the model through the GenAI SDK results in a 400 error:

ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'No embedding input is provided.', 'status': 'INVALID_ARGUMENT'}}

or if sent the direct HTTP request via the embedContent endpoint:

Publisher Model projects/.../locations/.../publishers/google/models/multimodalembedding@001 is not supported in the embedContent API.

A similar issue was raised previously: #1975

It was closed based on two points:

1. The model is deprecated
2. The GenAI SDK does not support multimodal embeddings

However, these assumptions no longer seem to apply:

• multimodalembedding@001 is still active until April 2027
• The GenAI SDK now supports multimodal embeddings (e.g. gemini-embedding-2-preview)

Would it be possible to support multimodalembedding@001 in the GenAI SDK, or provide guidance on the recommended migration path during this interim period?

[Venkaiahbabuneelam]

Hey @adubovik, Thanks for reaching out us!

gemini-embedding-2-preview, our first multimodal embedding model. It supports text, image, video, audio, and PDF inputs, mapping all modalities into a unified embedding space.
We recommend keeping an eye on our release notes for future updates.

Thanks

[adubovik]

@Venkaiahbabuneelam

Thank you for the suggestion — gemini-embedding-2-preview looks like a strong model. However, it is not equivalent to multimodalembedding@001, which our existing data pipelines depend on.

According to Google’s documentation, multimodalembedding@001 is expected to remain available until April 2027, while the Vertex AI SDK is scheduled for deprecation in June 2026. This creates a gap in supported access.

Could you clarify what the recommended approach is for using multimodalembedding@001 during this interim period?

While calling the API via raw HTTP is technically possible, it is not an ideal long-term solution compared to having proper SDK support.