OpenAI Models Head to Google Cloud's Model Garden - Three Models Including GPT-6 Astra, with Existing Cloud Commitments to Be Usable for Purchases

On October 21, 2026, OpenAI announced a partnership with Google Cloud under which its frontier models and Codex’s agent-style workflows will be offered in the model garden on Google Cloud¹. Google Cloud announced the same move in an official blog post dated October 22, saying it had started offering OpenAI models there².

According to Google Cloud, three models are involved, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, and they are going first to a limited group of enterprise customers as a private preview². OpenAI says general availability is planned within the next few weeks¹. Since the deals go through Google Cloud Marketplace, companies that have committed to a level of Google Cloud spending can count OpenAI model purchases against it.

What becomes available on Google Cloud

Google Cloud’s announcement describes the three models roughly as follows².

• GPT-6 Astra: for heavy work such as coding, operating computers, professional tasks and scientific research
• GPT-6.1 Sol: close to Astra in capability, at a lower cost
• GPT-6 Luna: a quick, economical model for routine work in high volume

Pricing and other details for each model are covered in our articles on GPT-6 Astra, GPT-6 Sol and Luna and GPT-6.1 Sol.

Customers admitted to the preview can call these models through Google Cloud’s APIs. Google Cloud adds that developers can also point Codex CLI or the Codex desktop app on their own machines at Google Cloud’s Gemini platform as the place where inference runs².

Google Cloud also writes that OpenAI models are to arrive later in Google Antigravity, Google’s agent-driven development environment, and in what it calls “Gemini for business.” OpenAI’s announcement names Antigravity and a product it calls “Gemini Agent” as upcoming destinations¹. Neither company gives a date.

One Google Cloud bill, and commitments count

For procurement teams, the key change is the contracting and payment path. OpenAI models are sold through Google Cloud Marketplace, and charges show up on the customer’s single Google Cloud invoice². Google Cloud says this spending reduces the balance of a customer’s existing commitment and also qualifies for Flexible Savings Plans (FSP), which apply across model providers. In effect, an organization can obtain the models under the Google Cloud agreement it already has instead of signing a separate contract with OpenAI.

OpenAI has announced a similar arrangement for Oracle Cloud. On June 10, 2026, it said Oracle Cloud customers would be able to apply eligible Oracle Universal Credits to OpenAI models and Codex³. The Google Cloud deal points the same way, letting companies buy OpenAI models through a cloud contract they already hold.

At the same time, OpenAI closes its announcement by stating that Microsoft is still its primary cloud partner¹.

How data and security are handled

Both announcements address companies in regulated industries. OpenAI describes the arrangement as combining its models with Google Cloud’s access controls, data governance and infrastructure¹.

Google Cloud states that what customers send and receive, including prompts, context and responses, stays inside the data boundary they choose and never goes into training base models². The models are hosted on infrastructure Google Cloud sets aside for them, and requests travel through isolated gateways. Each interaction with a model happens under an authenticated identity, and is covered by VPC Service Controls and by unified audit logs, according to the post.

Options for Zero Data Retention (ZDR), along with Confidential Compute, however, are described as coming soon and cannot be used yet². Organizations that require ZDR before adopting a model will have to wait for them.

What is still unknown

Neither announcement gives per-token prices, so whether the models will cost the same as on OpenAI’s own API is unknown for now. No regions are listed either, which leaves open which regions customers in Japan would be able to use. The general availability date is described only as “in the coming weeks,” with no specific day set¹.

Google Cloud presents the launch as making it the only hyperscale cloud that offers all three leading closed-source LLMs².

In practical terms, the news matters most to organizations that already have large Google Cloud commitments and want to use models inside that data platform and permission system: OpenAI models now become a procurement option for them. During the preview, though, access is limited to some enterprise customers, and neither prices nor regions have been published. Whether to fold OpenAI models into plans for spending down an existing commitment is a decision best made after seeing the prices at general availability and the timing of data protection options such as ZDR. Both companies provide a form for interested businesses.

Sources

1. OpenAI models are coming to Google Cloud - OpenAI official announcement (October 21, 2026)
2. Google Cloud expands model choice and flexibility, bringing OpenAI models to customers - Google Cloud official blog (dated October 22, 2026)
3. Access OpenAI models and Codex through your Oracle cloud commitment - OpenAI official announcement (June 10, 2026)