Published: 21 October 2026
On 21 October 2026, OpenAI and Google Cloud said GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are coming to Model Garden, Google Cloud’s model catalogue. Companies buy them through Google Cloud Marketplace, so the cost lands on their Google Cloud invoice and counts against money they have already committed to Google. For now, only selected large customers get access in a private preview, and OpenAI plans to open it to all in the coming weeks.
GPT-6 Astra, Sol and Luna join Gemini and Claude in Model Garden
Google describes Astra as the model for demanding work in coding, computer use and research. Sol comes close to Astra at a lower cost, and Luna is the fast, cheaper model for everyday tasks. Developers at preview customers can also connect Codex, OpenAI’s coding agent, to Google’s platform. From today, the Codex command line tool and desktop app can send their model requests to Google Cloud instead of OpenAI.
The models will later appear in Google Antigravity, Google’s coding tool, and in the Gemini Agent in Google’s business offering. Neither company gives a date for that step.
With this deal, Google says it is the only large cloud provider that offers all three leading closed models: its own Gemini, Anthropic’s Claude and OpenAI’s GPT models. Amazon Web Services has offered OpenAI models since April, after Microsoft’s exclusive arrangement ended, The Next Web and Reuters reported. OpenAI says Microsoft remains its primary cloud partner.
Buying GPT-6 through Google skips a separate OpenAI contract
The bigger change is in how you buy, not in the models. All use appears on one Google Cloud invoice, is paid from existing Google Cloud commitments and counts toward Google’s Flexible Savings Plans, which cover models from several makers. Google says customers can swap between models without separate vendor contracts.
Google also makes promises about data. It says prompts, context and outputs stay inside the customer’s own data perimeter and are never used to train models. The OpenAI models run on dedicated Google Cloud infrastructure. Every request runs under a known identity, with audit logs and VPC Service Controls, Google’s way of fencing off cloud data.
Do the OpenAI models on Google Cloud run in Europe?
Neither company says so yet. Dedicated hardware and a data perimeter do not tell you in which country your prompts are handled. For a European company, that is the first question to settle before any personal data goes in. Both announcements leave out these details:
- the regions where the models run, in Europe or elsewhere;
- prices, and whether a token costs the same as in OpenAI’s own API;
- zero data retention, which Google lists as coming soon;
- Confidential Compute, also coming soon;
- a date for general availability.
Google Cloud teams can now test GPT-6 without adding a supplier
In many companies, the buying process and the security review pick the AI model long before anyone compares answers. A company that runs on Google Cloud tends to use Gemini, because adding OpenAI means a new supplier, a new contract and a new review. This deal removes the separate contract and much of that review. It also makes a fair comparison possible: the same data, the same access rules and the same logs, with only the model changing. That is the comparison a team should run before it builds its agents or internal tools on one model.
For teams on Microsoft Azure and Microsoft 365, nothing changes. OpenAI still calls Microsoft its primary cloud partner, so Azure remains the obvious route for teams that already build there.
If you run on Google Cloud and held back on OpenAI models for contract reasons, sign up through Google’s interest form for OpenAI models. Then collect a set of real tasks from your own work to test with. Keep personal and secret data out of those tests until Google says where the models run and how long data is kept. Wait for prices before you plan to use them in daily work. If your developers already use Codex, switch it to Google only once you know where the requests are processed.
For more information, visit the official announcement of OpenAI models on Google Cloud on the Google Cloud blog.
