OpenAI's models are coming to a rival's shelf. Under a new partnership, OpenAI's frontier models and Codex agent workflows will be sold through Model Garden, the model catalogue in Google Cloud's AI platform, and billed through the Google Cloud Marketplace. General availability is planned within weeks, with an access request form open now.
The detail that matters most to buyers is about money rather than technology. Companies with Google Cloud spending commitments will be able to count purchases of OpenAI models against those commitments. For a large firm that has already promised Google a set amount of cloud spend, that removes the need for a separate contract, a new vendor review and another budget line before a team can use GPT models in production. OpenAI also pitches the deal at regulated industries, which get Google Cloud's access controls, central data governance and infrastructure around the models.
The second part reaches into Google's own agent products. OpenAI says its models will soon be available in Google Antigravity, the agent-first coding platform we review, and in the Gemini Agent. Antigravity has so far been built around Gemini models, including Gemini 4 Argon, so this gives its users a choice of model family inside the same tool. Google's side of the announcement frames it as offering customers a broad selection of models so they can pick the right one for each job, which is the same argument it already makes for hosting Claude and other third-party models on its platform.
For readers, the effect is mostly practical. Teams standardised on Google Cloud can try OpenAI models without leaving the console or the procurement process they already use, and Codex users may find it appearing in places they did not expect. It also shows how ordinary it has become for the big labs to sell through each other's clouds; model choice is turning into a feature that every platform has to offer.
The announcement leaves a lot unsaid. It names no specific models, gives no pricing and does not say whether rates or rate limits will match OpenAI's own API. Nor does it say how data handling will be split between the two companies, which is exactly what regulated buyers will ask first. Until those terms appear at launch, the deal is best read as a purchasing convenience rather than a new capability.