OpenAI Models Join Google Cloud's Model Garden With GPT-6 Astra

OpenAI and Google Cloud announced on 21 October 2026 that OpenAI's frontier models will be offered in Google Cloud's Model Garden, the catalogue where Google customers browse and deploy AI models. According to reporting by CNBC, the models named in the announcement are GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, and they sit next to Gemini in the same catalogue. OpenAI's Codex coding agent is included as well. The announcement was confirmed by both companies, and Ars Technica also reported it.

What was announced

The models are not yet open to every Google Cloud customer. According to Google, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out now in private preview, which means select enterprise customers get them first. OpenAI said in its blog post that general availability is planned in the coming weeks, but neither company has published a firm date, so buyers should not plan a launch date around one yet.

Google Cloud says it now counts as the only major hyperscale cloud that offers access to all three leading closed models. The Model Garden already lists more than 200 first-party, third-party and open models, according to Google Cloud's own description of the catalogue in 2026.

Why it matters for businesses that automate work

Until now, a company that wanted OpenAI models for automated work usually went through Microsoft Azure or OpenAI's own API. The brief for this story describes the change as ending an Azure-only arrangement for Google Cloud users. The reporting we checked does not support that wording in full. Microsoft remains OpenAI's primary cloud partner, according to the announcement coverage by The Next Web, and Amazon added OpenAI's models in April 2026. What is new is that Google Cloud customers can now license OpenAI models natively, inside their existing Google Cloud setup.

For routine work, the practical question is which model fits which job. Google and OpenAI describe the three models as different points on the cost and capability scale. GPT-6 Astra is built for demanding work such as coding, computer use and scientific research. GPT-6.1 Sol is described as close to Astra in capability at a lower cost. GPT-6 Luna is positioned as the fast, low-cost option for everyday work at volume.

What changes in practice for a business

The most concrete change is billing. According to Google Cloud's Michael Gerstenhaber, usage of the OpenAI models appears on a single Google Cloud invoice and can draw down existing cloud spending commitments. A company that already pays Google Cloud does not need a separate contract with OpenAI to try the models. That matters most for finance teams, who would otherwise have to add a new vendor, a new approval and a new payment method.

Data handling is the second change. Google said that customer prompts, context and outputs stay inside the customer's data perimeter and are not used to train models. The OpenAI models run on dedicated Google Cloud infrastructure, according to the same coverage. Owners should still ask their own IT or finance lead to confirm these terms in the contract before any customer or payroll data goes into a model.

Developers can also connect Codex command-line and desktop workflows to Google's Gemini platform for inference, starting on 21 October, according to reporting on the announcement. A small business without an in-house developer will rarely use this directly, but an outside developer or an automation partner may.

For a business that automates work, a sensible first step is to match one task to one model. Start with a low-risk job, such as sorting incoming invoices or drafting payment reminders for a person to approve. Compare the output and the cost of two models on the same sample before moving anything into live use.

Limits to keep in mind

A private preview is not a production service for most firms. The models may change before general availability, and the sources we checked did not publish per-use prices for each model, so the cost of a given workflow cannot be estimated from this announcement alone. Automation also does not remove the need for review. A model can misread a document, invent a figure or miss an exception, and a person still needs to check anything that moves money, changes records or reaches a customer.

What to watch next

Three things will tell buyers more in the coming weeks. The first is the general availability date and the published price for each model. The second is whether the models become available in every Google Cloud region, which none of the reporting has yet confirmed. The third is how Microsoft and Amazon respond on pricing and features, since both already sell OpenAI models. Until those details are published, the sensible approach is to test on a small set of tasks and wait for the price sheet before planning a rollout.