OpenAI Models and Codex on Google Cloud: What the Private Preview Includes
Technology News & Updates
Google Cloud and OpenAI announced on October 21, 2026, that OpenAI models and Codex workflows were coming to Google Cloud’s Model Garden. Google identified GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna for selected enterprise customers in private preview. The public documentation reviewed for this article still describes preview onboarding. The current guide also lists GPT-6.1 Terra. Teams should confirm access for their project before scheduling a migration.
How does access work?
Google Cloud’s Model Garden is the catalog where customers discover and enable models. OpenAI describes purchases through Google Cloud Marketplace and says eligible purchases can count toward existing Google Cloud spending commitments. The onboarding guide calls for project approval, billing and offer acceptance, model enablement and permissions for the calling identity.
The example in OpenAI’s guide uses a Google Cloud access token and a global publisher endpoint. A global endpoint does not establish that a workload satisfies a regional residency requirement. Teams with that requirement should confirm processing locations and contractual terms during onboarding.
What does the announcement say about Codex?
Google says preview customers can connect local Codex CLI and desktop workflows to its Gemini platform for inference. That describes a way to route model requests from those tools. Ask separately about any hosted execution, repository access or administrative features your team relies on.
Will an existing OpenAI integration work unchanged?
OpenAI documents text generation and streaming through the Responses API on Google Cloud. Its comparison also lists gaps, including Chat Completions, hosted web and file search, and server-side remote MCP. A team moving a document assistant should identify which capabilities it currently delegates to OpenAI before switching providers. Client-managed retrieval, for example, has different operational requirements from a hosted file-search feature.
Data handling deserves its own check. OpenAI’s guide lists default retention of up to 30 days and directs customers with Zero Data Retention requirements to confirm support during onboarding. Turning off stored responses in a request does not enable Zero Data Retention.
What should cloud teams evaluate first?
Choose one approved test project and one representative workflow. Compare authentication, permissions, feature coverage, data handling and billing with the current route. The preview is most immediately relevant to organizations that already manage workloads and procurement in Google Cloud.
Related on ASP.NET PB · OpenAI API integration in ASP.NET Core.
Comments
Post a Comment