OpenAI is bringing its frontier models to a rival hyperscaler. In a joint announcement on October 21, 2026, OpenAI and Google Cloud said three OpenAI models, GPT-6 Astra, GPT-6.1 Sol plus the lighter GPT-6 Luna, are now rolling out to selected enterprise customers inside Google Cloud's Model Garden, the catalogue Google uses to offer its own Gemini models alongside third-party and open alternatives. OpenAI's Codex agentic coding workflows are part of the arrangement, and the company expects general availability within a few weeks.
The commercial plumbing is the most concrete part of the deal. Customers will buy access through Google Cloud Marketplace, so usage lands on their regular Google Cloud invoice. Companies that have signed spending commitments with Google can count OpenAI usage against them, and Google says it also qualifies for its Flexible Savings Plans across model providers. For large buyers, that removes the need for a separate OpenAI contract when their AI budget already sits with Google.
Google's post describes the three models in its own terms. Astra is pitched for the hardest coding, computer use, professional and scientific work; Sol is presented as close to Astra's capability at a lower price; Luna is the fast, inexpensive tier for high-volume tasks. Developers can call the models through Google Cloud's APIs from today and route local Codex CLI and desktop sessions to the Gemini platform for inference. Both companies say OpenAI models will later appear in Google Antigravity, Google's agentic development environment, and in Google's business agent products.
Google is also making data-handling promises. It says customer prompts, context and outputs stay inside the customer's chosen perimeter and are not used to train foundation models, that the OpenAI models are hosted on infrastructure Google reserves for them, behind isolated gateways, and that every call is tied to an authenticated identity with audit logging and VPC Service Controls. Zero Data Retention options and Confidential Compute for these models are described as coming soon rather than available now.
With this addition, Google says it is the only hyperscale cloud offering all three of the leading closed-model families, a reference to its own Gemini line, Anthropic's Claude, which it already hosts, and now OpenAI. Thomas Kurian, Google Cloud's chief executive, framed the move as letting customers pick the right model for each job. OpenAI president Greg Brockman said it puts the company's most capable models and Codex where customers already build.
The announcement carries one line that matters for the wider market: OpenAI states that Microsoft remains its primary cloud partner. The deal therefore widens distribution rather than replacing Azure, and neither company disclosed pricing, revenue sharing or how much compute Google will dedicate. Reuters reported the agreement as part of OpenAI's broader push to sell more to enterprises.
Several practical questions remain open. The rollout is limited to a private preview for select customers, the governance features that regulated buyers often require are still marked as upcoming, and Google has not said which regions will host the models. How quickly enterprises shift spending will depend on those details once general availability arrives.