OpenAI models enter Google Cloud private preview
OpenAI and Google Cloud announced a partnership on October 21 to bring OpenAI models and Codex workflows into Model Garden. OpenAI’s announcement puts general availability in the coming weeks, making the current rollout an early access stage.
Google’s announcement names GPT 6 Astra, GPT 6.1 Sol and GPT 6 Luna for a private preview reaching selected enterprises. It says participants can use APIs and connect local Codex CLI and Desktop workflows to the Gemini platform for inference.
What the preview permits
The OpenAI access page sets a narrower boundary than a production launch. Access is by invitation for approved development and internal testing. The form explicitly excludes production use. Registering interest only requests consideration for early access.
That page lists model access through the Responses API on Google Cloud. It also lists Codex through the API in its app, command line interface and supported development environments. Teams evaluating the service therefore need both access approval and a use case within the preview’s stated limits.
Purchasing and data controls
OpenAI says model purchases will transact through Google Cloud Marketplace and can draw down Google Cloud spending commitments. It also says Microsoft remains its primary cloud partner. The announcement does not provide a price table or a regional availability matrix.
Google says prompts, context and outputs stay within designated data perimeters and are not used to train foundation models. It describes authenticated identities, VPC Service Controls and unified audit logging. These are the provider’s stated safeguards, rather than an independent security assessment.
Two additional protections are future features. Google labels Zero Data Retention options and Confidential Compute as coming soon. Its announcement gives model names but no configuration identifiers or exact general availability date.
Agent integration comes later
OpenAI models are also planned for Google Antigravity and the Gemini agent. OpenAI describes those integrations as arriving soon, without a specific date.
Google’s October 8 platform explanation distinguishes the Gemini agent from the model powering it. The agent supplies tools, context, skills and a workflow layer that can work across different models. That earlier architecture helps explain how model choice can change without replacing the surrounding agent.
Related reading covers the earlier Gemini workplace agent announcement.
Archival illustrative photograph by Wvdp, via Wikimedia Commons, dedicated to the public domain under CC0 1.0. It shows Google’s Eemshaven data center in January 2023, not a confirmed hosting location for this partnership.
