Google Cloud adds OpenAI models alongside Gemini and Claude

Google Cloud will now rent out OpenAI's models beside its own — and beside Anthropic's. The company has added three OpenAI models to its model garden, making it, in its own words, the only hyperscale cloud provider offering Gemini, Claude and GPT on a single platform. ¹

The three models are GPT-6 Astra, aimed at heavy work such as coding, computer use, professional tasks and scientific research; GPT-6.1 Sol, pitched as near-Astra intelligence at a lower price for agentic coding and professional work; and GPT-6 Luna, a fast, cheap option for everyday jobs at scale. ¹

On paper, the appeal is less about any single model than about not having to pick a side. Google says customers can swap between models on performance, latency or cost "without architectural lock-in or separate vendor contracts". ¹

One cloud, one bill

The commercial detail matters as much as the model list. OpenAI models are transacted through Google Cloud Marketplace, so usage lands on a single Google Cloud invoice, draws down existing cloud commitments and counts towards Flexible Savings Plans that already span multiple model providers. ¹

That is a quiet but significant change for enterprise buyers. Until now, adding a frontier model from a second vendor usually meant a second contract, a second billing relationship and a second set of security reviews. Google is betting that procurement friction, not model benchmarks, is what keeps customers from mixing providers.

The rollout is deliberately cautious. The three models are in private preview for selected enterprise customers, with an interest form open for others, and Google says customers can already connect their local Codex CLI and Desktop workflows to its Gemini platform for inference. ¹ Availability inside Gemini for Business and Google Antigravity is promised "coming soon". ¹

The security pitch

Google is also trying to answer the obvious objection: why would a company route confidential prompts through a rival's model on Google's hardware? The company says third-party and OpenAI workloads stay inside a customer's designated data perimeter, are never used to train foundation models, and have Zero Data Retention options coming soon. ¹

It adds that OpenAI models run on dedicated Google Cloud infrastructure, with Confidential Compute due later, and that every model and agent interaction runs under an authenticated identity with VPC Service Controls and unified audit logging behind it. ¹

Box, an early named customer, framed the arrangement around that governance layer. "By using OpenAI models through Google Cloud, we will be able to give customers access to OpenAI's frontier AI models alongside Google Cloud's security and compliance capabilities," said Yashodha Bhavnani, Box's VP of AI Products. ¹

The awkward part

There is a tension Google would rather not dwell on. Every GPU-hour sold to serve a rival's model is capacity that does not go to its own Gemini team, and the same engineering effort strengthens the competitors it is trying to beat. The Decoder notes that this resource conflict has been reported as a factor in researcher departures, including Gemini co-lead Noam Shazeer's move back to OpenAI. ²

Google's counter-argument is that openness wins the platform war. If enterprises are going to run several models anyway, the company would rather be the counter they buy them from than lose the workload entirely to AWS or Azure.

Our opinion

Google has made the pragmatic call, and it is the right one commercially even if it feels odd culturally. The premium in enterprise AI has moved from "whose model is best this quarter" to "who can govern, bill and audit many models without the customer building a second IT department". Being the marketplace for your own rival is an admission that no single model wins every workload, and Google gets paid either way.

The risk is strategic drift. A cloud business that profits from OpenAI and Anthropic traffic has a weaker incentive to make Gemini the default, and internal teams notice when the platform treats a competitor's model as a first-class citizen. The private preview, the missing Antigravity integration and the "coming soon" caveats suggest Google is testing demand before it commits to how visible that choice really becomes. Buyers should judge the offer on three things the blog does not settle: how pricing compares with going direct, whether the ZDR and Confidential Compute promises arrive on time, and whether model choice survives contact with Gemini for Business, where the default will be hard to resist.