Hi there,
I’ve been using Gemini 2.5 Pro for a couple of months now in BigQuery, using it to categorise product names, descriptions and tidy up meta data.
I want to use Gemini 3 Pro, but I can’t seem to access it. Am I missing an additional step of config?
My update query is very simple, looks like this (you can see how easily I’ve updated it previously)
CREATE OR REPLACE MODEL
`{project-id}.ml_models.gemini-fl`
REMOTE WITH CONNECTION `projects/{project-id}/locations/eu/connections/vertex_ai_connection`
-- OPTIONS (ENDPOINT = 'gemini-2.5-flash-lite');
-- OPTIONS (ENDPOINT = 'gemini-2.5-pro');
OPTIONS (ENDPOINT = 'gemini-3-pro-preview');
I keep hitting this error:
Not found: Publisher Model `projects/(PII Removed by Staff)/locations/europe-west4/publishers/google/models/gemini-3-pro-preview` not found.
What am I missing?
Thanks folks!
And… when I try setting up a new model, it’s just not in the dropdown options, any help would be hugely appreciated, thanks!
Hi Joe- For now, you cannot use the shorter model name for Gemini 3.0 Pro Preview but will need to provide the full endpoint in the argument. Here’s the section of the docs for AI.GENERATE that specify how to use Gemini 3.0 Pro Preview. I just tested it with the CREATE MODEL statement and using that model with AI.GENERATE_TEXT, and that works as well.
Full endpoint will look like this (swapping in your own PROJECT_ID):
https://aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/global/publishers/google/models/gemini-3-pro-preview
Let me know if that works!
2 Likes
Hi Alicia,
Thanks for the snappy response on this, full end-point approach totally makes sense.
I’ve adapted the query:
CREATE OR REPLACE MODEL
`trade-products-handling.ml_models.gemini-fl`
REMOTE WITH CONNECTION `projects/PROJECT_ID/locations/eu/connections/vertex_ai_connection`
OPTIONS (ENDPOINT = 'https://aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/global/publishers/google/models/gemini-3-pro-preview’);
Then I hit this error:
Permission denied: Permission denied on resource project {project_id}..
I’ve tried to amend the endpoint so it serves from EU:
'https://aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/eu/publishers/google/models/gemini-3-pro-preview’);
This takes a while longer to process which makes me think it’s get further in the job and then this error occurs:
Prediction service is unreachable. Please check the endpoint location and retry later.
So I feel like this may be a location issue.
Cheers!
I just tested creating a remote model to Gemini 3.0 Pro Preview within an EU multiregion dataset using the global endpoint, and was able to run a prompt-only query without a Permission denied error. Perhaps try this first to rule out other data permission issues?
SELECT *
FROM
AI.GENERATE_TEXT(
MODEL `eu_data.gemini_3_pro_eu`,
(SELECT 'What is the purpose of dreams?' AS prompt));
Because this is a global endpoint, it doesn’t adhere to the DRZ, so bear that in mind. I don’t think the endpoint is yet available from the EU region specifically (Deployments and endpoints | Generative AI on Vertex AI | Google Cloud Documentation)
