OpenAI Puts GPT-6 Astra, Sol and Luna on Google Cloud in Deal Between Rivals
OpenAI will sell its frontier models and its Codex coding agent through Google Cloud, with GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna entering private preview in Model Garden on October 21. The arrangement gives OpenAI a third major cloud venue after Microsoft Azure and Amazon Bedrock, and turns two companies chasing the same enterprise budgets into partners.
Two rivals sign the same page
On Wednesday, October 21, OpenAI said its frontier models and its Codex coding agent will be sold through Google Cloud — a striking arrangement between two companies that compete for the same enterprise AI budgets.
What is actually arriving
Three models lead the rollout: GPT-6 Astra, the flagship built for demanding coding, computer-use, professional and scientific research workloads; GPT-6.1 Sol, which OpenAI and Google Cloud position as near-Astra intelligence at a lower price across agentic coding and professional work; and GPT-6 Luna, the cheaper tier.
All three began rolling out in private preview inside Google Cloud's Model Garden on the day of the announcement, open to a select group of enterprise customers. Billing runs through Google Cloud Marketplace, meaning existing Google Cloud commitments can be spent on OpenAI models. Codex is part of the package. General availability is planned for the coming weeks, not quarters.
OpenAI also said the models will appear in Google Antigravity, Google's agentic development platform, and in the Gemini Agent, which routes tasks across model families.
Why Google would host its main rival
Google's calculus is straightforward: enterprises increasingly refuse to standardize on a single model family. Google Cloud has spent the year arguing that Antigravity and Gemini Enterprise should be a marketplace of models rather than one vendor's stack. On the same day, Google Cloud confirmed that Anthropic's Claude Opus 5.5 and Sonnet 5.5 are now selectable in Antigravity for paid Google AI Pro and Ultra users.
Gemini Agent already routes work across Gemini and Claude models, with OpenAI and open-weight models slated to follow. Adding OpenAI closes the loop.
For OpenAI, the deal buys distribution inside accounts it has struggled to reach: organizations that keep their data, spending and governance anchored on Google Cloud. Reuters reported the move as an enterprise push, framing it as evidence that large cloud providers are becoming the key distribution channel for frontier AI.
The Microsoft question
The arrangement is possible because OpenAI's ties to Microsoft loosened. Microsoft gave up exclusive access to OpenAI's technology in April 2026, freeing OpenAI to sell through Amazon and Google while remaining OpenAI's primary cloud partner through 2032.
OpenAI models already run on Microsoft Azure and Amazon Bedrock. Google Cloud becomes the third major venue, and the deal itself was finalized in May after months of discussion.
What to watch
Two things. First, pricing and rate limits in Model Garden, which will determine whether companies actually migrate their workloads or simply experiment on the side.
Second, latency and feature parity. A model served through a partner cloud is rarely identical to the first-party version, and enterprise buyers comparing evaluations will notice the gap. OpenAI's own published numbers for GPT-6.1 Sol, for instance, put it within about two points of Astra on OSWorld 2.0 at maximum reasoning effort — a margin that small makes serving details matter as much as the model card.
There is also the question of attention. Google is simultaneously shipping its own Gemini 4 Argon, hosting Anthropic's Claude 5.5 family, and now reselling OpenAI's top tier. The pitch is model choice; the risk is that buyers treat the whole catalogue as interchangeable compute and buy on price alone.
For now, the headline is simpler than the strategy: the company that built GPT-6 will rent it out from the company that built Gemini.