OpenAI models on Google Cloud: GPT-6 Astra, 6.1 Sol and Luna enter private preview, and what it means for Microsoft

The OpenAI Google Cloud deal, announced on October 21, 2026, brings OpenAI's models to the cloud of its most direct rival. Google Cloud says GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are "rolling out now in private preview" in its Model Garden for select enterprise customers, who pay through Google Cloud Marketplace. OpenAI's own post adds Codex and says general availability is "planned in the coming weeks." With it, OpenAI's models are sold on all three of the big US clouds: Microsoft Azure, Amazon Bedrock and now Google Cloud.

This story is based on OpenAI's and Google Cloud's announcements, the April 27, 2026 OpenAI–Microsoft amendment, AWS and Microsoft availability posts, and reporting by Axios, SiliconANGLE and Reuters, checked October 22, 2026; the announcements, availability posts and status were re-checked October 23, 2026. For what each GPT-6 model does and costs on OpenAI's own API, see our GPT-6 launch guide.

The short version

• What: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in Google Cloud's Model Garden, plus a way to point the Codex CLI and desktop app at Google for inference.
• Status: private preview for select enterprise customers, by sign-up form (Google). OpenAI says GA is planned "in the coming weeks."
• Billing: through Google Cloud Marketplace; usage counts against existing Google Cloud commitments.
• Not disclosed: prices, regions, a GA date and any financial terms.
• Microsoft: still OpenAI's "primary cloud partner," but its license has been non-exclusive since April 27, 2026.

Which OpenAI models are on Google Cloud?

Three, according to Google Cloud's October 21 post by Michael Gerstenhaber, its vice president of product management for Gemini:

• GPT-6 Astra, for "demanding work" in coding, computer use, professional tasks and scientific research;
• GPT-6.1 Sol, which Google describes as near-Astra intelligence at lower cost;
• GPT-6 Luna, a fast, cheaper model for everyday work at scale.

Google says more are "coming soon." OpenAI's post names no models at all. It speaks of "frontier models" and Codex, and adds that OpenAI models will come "soon" to Google Antigravity, Google's coding tool, and Gemini Agent. Google's post names Gemini for business and Antigravity.

How do you get access, and what does it cost?

As of October 23, 2026, by asking. Both companies link to an interest form, and Google says the preview is limited to "select enterprise customers." Once in, customers call the models through Google Cloud APIs, and can connect local Codex CLI and desktop workflows to Google's Gemini platform for inference.

The money runs through Google. Purchases go through Google Cloud Marketplace, appear on a single Google Cloud invoice, draw down existing cloud commitments and count toward Flexible Savings Plans across model providers, Google says. Neither company lists per-token prices or regions. On Bedrock, by comparison, AWS said in June that OpenAI pricing "matches OpenAI first-party rates."

On data, Google says prompts, context and outputs stay within the customer's "designated data perimeter" and aren't used to train foundation models. The models run on dedicated Google Cloud infrastructure behind isolated gateways. Zero data retention and Confidential Compute are both "coming soon."

Where are OpenAI models available now?

Last verified: October 23, 2026

Microsoft's GPT-6 Astra post calls the model "generally available for all customers in Microsoft Foundry," while the same page's summary says it "begins rolling out through the Microsoft Foundry Limited Access Program." The table uses the body text.

What does the deal mean for Microsoft?

It's the second half of a change Microsoft agreed to in April. On April 27, 2026, OpenAI and Microsoft amended their partnership, as covered in our OpenAI explainer. OpenAI's post on the amendment says:

• Microsoft's license to OpenAI IP for models and products, which runs through 2032, "will now be non-exclusive";
• "OpenAI can now serve all its products to customers across any cloud provider";
• Microsoft "remains OpenAI's primary cloud partner," and OpenAI products "will ship first on Azure, unless Microsoft cannot and chooses not to support the necessary capabilities";
• Microsoft stops paying OpenAI a revenue share, while OpenAI's payments to Microsoft continue through 2030 at the same percentage, subject to a total cap;
• Microsoft remains "a major shareholder."

AWS moved first. It opened a limited preview of OpenAI models on Bedrock on April 28, the day after the amendment, and has since made GPT-5.5, GPT-6 Astra and GPT-6.1 Sol generally available. Google's preview arrives almost six months later. OpenAI's October 21 post repeats that Microsoft remains its primary cloud partner. Neither announcement includes comment from Microsoft, and Axios's report describes no change to Azure terms; it credits the April changes with opening the way to broader distribution.

What Microsoft loses is exclusivity, not access: it keeps its license and the right to ship OpenAI products first. What changes is competition for the same enterprise budget. Google Cloud now says it's the only hyperscale cloud offering Gemini, Claude and GPT models together. Anthropic already sells Claude on all three clouds, as our revenue-accounting analysis notes. Mike Leone of Moor Insights & Strategy told Axios that "Gemini now has to win those customers with OpenAI sitting right next to it."

How big is the OpenAI Google Cloud deal?

Not disclosed. Neither OpenAI's nor Google's post gives financial terms, Axios's and SiliconANGLE's reports give none, and Axios noted that neither company has put a number on the earlier AWS arrangement either. We found no named outlet reporting a deal value as of October 23, 2026.

Does OpenAI use Google Cloud for its own compute?

That is a separate arrangement, and older. In June 2025, Reuters reported, citing three sources, that OpenAI had agreed to add Google Cloud capacity for training and running its models, a deal finalized in May 2025. In July 2025, The Information reported that OpenAI had begun renting Google's TPUs, as relayed by DatacenterDynamics. Reuters then reported that OpenAI said it was testing some TPUs but had no plans to deploy them at scale, according to a GuruFocus summary on Yahoo Finance. Neither company's October 21 post mentions compute.

What's confirmed, reported and uncertain

• Confirmed (OpenAI, Google Cloud, October 21, 2026): the three models, private preview, Marketplace billing, use of existing commitments, Codex inference through Google, plans for Antigravity and Gemini Agent, and Google's data-handling terms.
• Conflicting: OpenAI describes GA as "planned in the coming weeks" without mentioning a preview; Google says the models are in private preview now. Both can be true, but no GA date exists.
• Confirmed (OpenAI, April 27, 2026): the non-exclusive license, any-cloud rights, "primary cloud partner" and ship-first-on-Azure terms.
• Reported: OpenAI's 2025 compute deal with Google Cloud (Reuters) and its TPU use (The Information).
• Uncertain: prices, regions, the GA date, whether more OpenAI models follow, and any deal value. If you're choosing between clouds for the same model, our model-routing explainer covers how teams mix providers.