Model not available in us-central1 despite working locally with global location

Google Cloud Build with AI Agents
1

Hi Team,

I’m working with an ADK agent using the gemini-3-flash-preview model.

Setup:

  • Locally, I configured the environment as:
GOOGLE_GENAI_USE_VERTEXAI=TRUE
GOOGLE_CLOUD_LOCATION=global
GOOGLE_CLOUD_PROJECT={project_id}

LEGAL_AUDIT_MODEL=projects/{project_id}/locations/global/publishers/google/models/gemini-3-flash-preview
  • With this setup, everything works correctly in local testing.

Issue:

  • When deploying the agent to Agent Engine, I’m required to set the runtime location to us-central1.
  • However, after deployment, I consistently receive an error indicating that the model is not available in us-central1, even when I explicitly provide the full global model path:
projects/{project_id}/locations/global/publishers/google/models/gemini-3-flash-preview

Questions:

  1. Does Agent Engine restrict model usage strictly based on the deployment region?
  2. Is it expected that models available in global cannot be invoked from a regional runtime like us-central1?
  3. What is the recommended approach to use preview/global models (like gemini-3-flash-preview) with Agent Engine?
  4. Are there any workarounds (e.g., forcing global routing or using a compatible regional model)?

Any guidance or clarification would be really helpful.

Thanks!

2

Closing the loop on this.

The issue has been resolved. Sharing the GitHub discussion for anyone facing something similar: Cannot Use Gemini 3 with Agent Engine · Issue #3628 · google/adk-python · GitHub