OpenAI's GPT-6 models are coming to Google Cloud, putting ChatGPT's maker on all three big U.S. clouds

OpenAI and Google Cloud said on Wednesday, Oct. 21, 2026, that OpenAI's frontier models and its Codex coding agent are coming to Google Cloud's Model Garden, with GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna rolling out now in private preview to select enterprise customers and general availability planned in the coming weeks.

Six months ago OpenAI's models lived on one cloud. Now they're on all three, and the newest landlord is the company building the chatbot most likely to eat ChatGPT's lunch. For Google, it's a confident move: let customers pick GPT-6 inside Google's own console, collect the cloud bill either way, and bet that Gemini wins on its own merits. For OpenAI, it's the enterprise-sales shortcut: big companies already have Google Cloud budgets committed, and now those dollars can buy OpenAI. The losers are anyone hoping model choice would keep the clouds honest. Every model is now everywhere; the cloud contract is the real lock-in.

OpenAI's frontier models, and Codex, its coding agent, are coming to Google Cloud's Model Garden, the catalog of models a customer can call from Google Cloud. Google describes the three models in the private preview. GPT-6 Astra is for demanding coding, computer use, and professional and scientific work. GPT-6.1 Sol offers intelligence close to Astra's at a lower cost. GPT-6 Luna is for fast, cheaper everyday work. A private preview is a limited release to selected customers, not a public launch. Google says that preview is open now to select enterprise customers. Those descriptions are Google's.

Starting today, Google says, those customers can build with the models through Google's APIs, the doors a program uses to ask for an answer. They can also connect the Codex command-line tool and the Codex desktop app, running on their own machines, to Google's Gemini platform for inference, the step where a trained model produces the answer. OpenAI says general availability, a release a customer can buy without a private invitation, is planned in the coming weeks. It is not generally available today. Those lines are Google's and OpenAI's.

The bill runs through Google. Customers buy the models on Google Cloud Marketplace, see them on one Google Cloud invoice, and can draw the cost down from Google Cloud spending they have already committed. Google says that spending also counts toward Flexible Savings Plans, its discount for committed cloud use. Those lines are Google's. Neither company published a price or a revenue split.

Coming soon, and not part of today's preview, Google says: OpenAI models inside Google Antigravity, Google's coding tool, and inside Gemini for business and the Gemini Agent.

Google's data claims, as Google states them. A customer's prompts, the extra context sent with them, and the model's outputs stay inside that customer's data perimeter, the boundary around that customer's own cloud setup, and are not used to train foundation models. A foundation model is the large model underneath a product. OpenAI's models run on dedicated Google Cloud infrastructure, machines set aside for this service. Google lists Zero Data Retention, an option not to keep the prompt and the reply after the request, and Confidential Compute, a mode where the cloud operator cannot read the job, as coming soon. Those options are not part of the preview described today. Those lines are Google's.

Google says it is now the only hyperscale cloud, one of the few companies that rent computing at that size, offering all three leading closed-source large language models, alongside more than 200 other models. Closed-source means the company does not publish the model's weights, the numbers that make it work. OpenAI says Microsoft remains its primary cloud partner. Those lines are Google's and OpenAI's.

Greg Brockman, OpenAI's president and co-founder, said the partnership makes "our most capable models and Codex available where customers are already building." Thomas Kurian, Google Cloud's chief executive, said customers can "choose the right model for any given job." Those quotations are theirs.

Reuters and CNBC set the sequence around the deal. OpenAI's exclusive arrangement with Microsoft ended in April, and Amazon Web Services began hosting OpenAI models right after that. Google is the third big U.S. cloud. OpenAI and Google also compete directly in chatbots and in search. That sequence is Reuters's and CNBC's. It is context around today's announcement, not a term of the contract either company printed.

In plain terms, OpenAI and Google said on Wednesday that GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, plus the Codex coding agent, are in private preview for selected enterprise customers on Google Cloud. A wider release is planned in the coming weeks. Customers can pay with Google Cloud commitments they already have. Neither company published a price. The models are not on sale to everyone today.