On October 21, 2026, Google Cloud announced something that would have sounded odd a couple of years ago: it is now selling OpenAI's models. GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are arriving in Google Cloud's Model Garden, the catalog where Google also lists its own Gemini models and a long tail of third-party ones. It is a private preview for selected enterprise customers, not a public launch, so most readers cannot use it today.
Still, the move matters even if you never open a Google Cloud console. It shows how the biggest AI labs are changing how they sell: instead of one lab, one cloud, one contract, the same model can now be bought through several clouds on the same invoice. This post walks through what was actually announced, what was left out, and what to watch next. Everything below comes from the announcement coverage cited inline; where reports differ or a detail was not disclosed, we say so.
Quick summary: On October 21, 2026, Google Cloud added three OpenAI models (GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna) to its Model Garden as a private preview for select enterprise customers. Billing runs through Google Cloud Marketplace on the same invoice, and usage counts toward existing Google Cloud commitments. Zero Data Retention and Confidential Compute are listed as coming soon. Pricing and a firm general-availability date were not disclosed; OpenAI said general availability is planned in the coming weeks.
What Google Cloud actually announced
Three OpenAI models are entering the preview, each pitched for a different job. GPT-6 Astra is positioned for demanding coding, computer-use, professional and scientific research tasks. GPT-6.1 Sol is described as close to Astra in intelligence but cheaper to run, aimed at agentic coding, computer use and professional work. GPT-6 Luna is the fast, cost-efficient option for everyday workloads at scale (Source: TestingCatalog).
Access is limited. The models are rolling out in a private preview for select enterprise customers, Google Cloud says more customers will get access later, and interested organizations apply through an interest form (Source: TestingCatalog). Preview participants can call the models through Google's APIs, and they can connect local Codex CLI and Codex Desktop workflows to Google's platform for inference (Source: Technobezz).
The models sit in Model Garden, which the same report says also contains more than 200 other models (Source: TestingCatalog).
The buying experience: one invoice, existing commitments
The most practical detail is not the models but the billing. OpenAI usage is purchased directly through Google Cloud Marketplace, and charges appear on the same Google Cloud invoice. Usage applies toward a company's existing cloud commitments and counts toward Flexible Savings Plans across model providers (Source: TestingCatalog). For a large company that has already promised Google a multi-year spend, that is a big deal: it can put OpenAI models on a budget line that is already approved, with no separate procurement cycle.
On data handling, coverage says prompts, context and responses stay within the customer's designated data perimeter and are not used to train foundation models (Source: Technobezz). Two protections that regulated industries usually ask for are not in the first release: Zero Data Retention options and Confidential Compute are both listed as coming soon (Source: TestingCatalog).
| Item | Status at announcement | Source |
|---|---|---|
| Models | GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna | TestingCatalog, Technobezz |
| Access | Private preview, select enterprise customers, interest form | TestingCatalog |
| Billing | Google Cloud Marketplace, same invoice, counts toward commitments | TestingCatalog |
| Zero Data Retention | Coming soon | TestingCatalog, Technobezz |
| Confidential Compute | Coming soon | TestingCatalog, Technobezz |
| Antigravity and Gemini for business | Coming soon | TestingCatalog, Technobezz |
| General availability | Planned for the coming weeks; no date given | TechEnclave |
| Pricing | Not disclosed in the coverage reviewed | TechEnclave |
What is still missing
The coverage is explicit about gaps. OpenAI models inside Google Antigravity (Google's coding environment) and inside Gemini for business are described as coming soon rather than live (Source: Technobezz). Pricing, regional availability, usage limits and model-specific specifications were not stated in the TechEnclave write-up, and there is no final general-availability date, only the intention to open up in the coming weeks (Source: TechEnclave). The same report notes that Microsoft remains OpenAI's primary cloud partner, so the Google deal adds a route rather than replacing an existing relationship (Source: TechEnclave).
- Confirmed: three named OpenAI models, private preview, Marketplace billing, data kept in the customer's perimeter and not used for training.
- Coming soon: Zero Data Retention, Confidential Compute, and OpenAI models in Antigravity and Gemini for business.
- Not disclosed: per-token pricing, a regional availability list, usage limits and a hard general-availability date.
Why this is a bigger shift than it looks
For years the story was exclusivity: a lab's best model lived on a lab's preferred cloud. A multi-cloud sales approach flips the logic. When the same GPT-6 family is available on more than one platform, buyers can pick a cloud for security, billing or latency reasons and then pick a model for quality, without being forced to take the model that ships with the contract. It also gives each cloud a reason to keep its catalog broad. Google gets to say customers do not have to leave its platform to try a rival's frontier model, and OpenAI gets another sales channel at the moment enterprises are deciding where to standardize.
There is a flip side. A catalog with many models is only useful if you can compare them on your own tasks, and most teams still pick a model by habit. Listing three OpenAI models next to Gemini and Claude does not tell you which one is cheaper per finished task, which refuses fewer harmless requests, or which holds up on long documents. Those answers still come from testing.
What this means if you are not an enterprise buyer
Nothing changes for your ChatGPT, Gemini or Claude subscription today. The preview is a business-to-business channel, and the sources we reviewed do not describe any consumer-facing change. What you can do is use the news as a prompt to test models side by side before your company or team commits to one. Two practical questions are worth asking any vendor in this space: which data-protection features are live right now versus coming soon, and what the true cost per completed task is, not just the sticker price per million tokens.
When a vendor says a protection is 'coming soon,' treat it as absent for planning purposes. Write down which of your requirements (retention, confidential compute, regional hosting) are live today, and re-check them when general availability is announced.
FAQ
01Which OpenAI models did Google Cloud add?
GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, announced on October 21, 2026, in Google Cloud's Model Garden.
02Can anyone use them now?
No. It is a private preview for select enterprise customers, and interested organizations apply through an interest form. General availability is described as planned for the coming weeks, with no date given.
03How is it billed?
Through Google Cloud Marketplace on the same Google Cloud invoice. Usage can count toward existing cloud commitments and Flexible Savings Plans. Per-token pricing was not disclosed in the coverage we reviewed.
04Does this replace OpenAI's Microsoft relationship?
No. TechEnclave reports that Microsoft remains OpenAI's primary cloud partner and the Google deal adds another route.
05Is my data used to train the models?
Coverage says prompts, context and responses stay within the customer's data perimeter and are not used to train foundation models. Zero Data Retention and Confidential Compute are listed as coming soon.
If you would rather compare GPT, Claude, Gemini and other leading models yourself than wait for an enterprise cloud contract, LumiChats lets you run the same prompt across 40+ premium AI models in one subscription, so you can see which one earns your work.
