Google AI Studio key fails with Mosaic AI Model Serving through Vertex AI provider

Use Mosaic AI Model Serving with custom provider and OpenAI-compatible endpoint.

Written by kaushal.vachhani

Last published at: July 10th, 2025

Problem

You’re attempting to serve Gemini models through Mosaic AI Model Serving using the built-in Google Cloud Vertex AI provider. You use a Google AI Studio-generated API key in your configuration, and authentication fails with a generic internal error.

("error_code": "INTERNAL_ERROR", "message": "Internal Error"}

 

Cause

You’re trying to use the Google Cloud Vertex AI provider with a Google AI Studio API key.

 

The Google Cloud Vertex AI provider in Mosaic AI Model Serving is designed to work specifically with Google’s service account keys. 

 

Solution

To serve a Gemini model continuing to use a Google AI Studio API key, set up the external model in Mosaic AI Model Serving with the Custom Provider configured in an OpenAI-compatible format.

 

Use the following configuration. 

Provider: Custom Provider

Custom Provider Model URL: https://generativelanguage.googleapis.com/v1beta/openai/chat/completions

Custom Provider Authentication Type: Bearer Token Authentication
  
Bearer Token: {{secrets/scope-name/gemini-api-key}}  # Use secrets!

Model Name: <Your gemini model> #gemini-2.0-flash

 

For further information on custom provider setup, refer to the “Custom provider” section of the External models in Mosaic AI Model Serving (AWSAzureGCP) documentation. You can also refer to the “Google Cloud Vertex AI” section of the same documentation.

 

For information on Google Gemini OpenAI compatibility, refer to the Google OpenAI compatibility documentation.