Google Cloud announced on October 22 that OpenAI models are now available in its Model Garden. The first three are GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. They open as a private preview to selected enterprise customers, and companies that want access must first submit an application form. OpenAI also published a post the same day confirming the partnership.
As Google describes them, the three models split the work as follows. Astra targets the hardest part of coding, computer use, professional tasks and scientific research. Sol comes close to Astra in intelligence at a lower cost, and is pitched at agentic coding and professional work. Luna is built for speed and cost efficiency, for running everyday tasks at scale.
How billing and spend commitments work
The most significant part of the deal is how the models are purchased. OpenAI models are sold directly through Google Cloud Marketplace, with usage charged to the same Google Cloud bill. That spend can be applied against cloud commitments an enterprise has already signed, and it also counts toward Flexible Savings Plans that span model providers. For large customers already holding multi-year committed spend with Google Cloud, using GPT no longer requires a separate OpenAI contract.
On the developer side, customers can call the models through the API now. Local workflows built on the Codex CLI and the Codex desktop app can also be pointed at the Google Cloud platform for inference. Google lists OpenAI models in the Google Antigravity coding tool and in Gemini Enterprise as "coming soon," without giving a date.
On data handling, Google says prompts, context and outputs stay within the data boundary the customer specifies and are not used to train foundation models. The models run on dedicated Google Cloud infrastructure, requests are routed through an isolated gateway, and third-party model providers have no authorized access to the customer's cloud environment. The zero data retention option and confidential computing are not live yet. The announcement says only that they are coming, and the text does not say which regions will be supported.
Michael Gerstenhaber, Google Cloud's VP of product management for Gemini, said in the announcement:
"Today, we are expanding that foundation by launching OpenAI models in Google Cloud's model garden."
OpenAI's cloud footprint keeps loosening
Put OpenAI's cloud deals of recent years side by side and this step is hardly a surprise. After Microsoft invested in 2019, Azure was for a long time the only public-cloud outlet for OpenAI models. When the Stargate project was announced in January 2025, the two sides replaced exclusivity with a right of first refusal for Microsoft, and from then on OpenAI could buy compute from other providers.
In June of that year, Reuters reported that OpenAI had added Google Cloud to its list of compute suppliers, at which point Google was the party selling GPU and TPU capacity. In August, OpenAI's open-weight model gpt-oss arrived on Amazon Bedrock, the first time an OpenAI model appeared on AWS. With this latest move, closed flagship models are now in Google's own model marketplace, sold alongside Gemini.
On Google's side, the announcement opens by stressing that it is the only hyperscaler offering all three closed frontier models: Gemini, Claude and GPT. For Google, when a customer picks GPT on its platform, the money still flows through the Google Cloud bill. OpenAI's announcement continues to call Microsoft its primary cloud partner. Microsoft has not responded publicly, and there is no word of any change to the terms of OpenAI services on Azure.
Sources: Google Cloud official blog, OpenAI official announcement, Reuters, CocoLoop; used to verify the names of the first models, the scope of the private preview and how billing credits apply.