Google Cloud makes xAI's Grok 4.7 generally available on two endpoint options

Google Cloud announced general availability for Grok 4.7 on two endpoint options.

Google Cloud marked xAI's Grok 4.7 generally available on October 27 through Gemini Enterprise Agent Platform. The release lists production use on the global endpoint and the US multi-region endpoint. Grok 4.7 was already available through xAI's own console. The October change is a new cloud availability milestone for an existing model. [Google Cloud release notes](https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes#October_27_2026) · [Google Cloud model documentation](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/grok/grok-4-7) · [xAI console changelog](https://x.ai/api/changelog)

Why the route matters

For a developer, the release can change how an application obtains a model without changing which model it asks for. An organization using Google Cloud can evaluate the partner model alongside its existing deployment controls. Google's documentation identifies it as grok-4.7. The route has its own service details, so teams should compare the documented interfaces and response behavior before assuming that code written for another provider endpoint will run unchanged. [Google Cloud model documentation](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/grok/grok-4-7)

The endpoint choice also needs care. A US multi-region endpoint and a global endpoint describe available service routes. An organization with a contract or policy about data location still needs to check the terms for the endpoint and feature it intends to use. The product's general availability label applies to the release Google described. Individual capabilities documented for the partner model can have separate stages.

What to compare in a migration

A production trial can replay representative requests with permitted test data and compare output structure, errors and end-to-end latency. If an application uses tool calls, the trial should include malformed tool responses, timeouts and permission denials. That helps distinguish differences in model behavior from differences in the integration around it. It also lets operators see whether an existing fallback would respond correctly when a request fails.

The Google Cloud release gives teams an additional way to access the model. Its value depends on whether that route meets the application's deployment requirements and maintains the behaviors users already rely on.