23 OCT 2026 — OpenAI's newest models are coming to Google Cloud, the last of the three big US cloud platforms to carry them. Google said GPT-6 Astra, GPT-6 Luna and GPT-6.1 Sol are rolling out in private preview to selected enterprise customers in its model garden, the catalogue of models on its Gemini Enterprise Agent Platform. OpenAI's news feed timestamps its own post at 12:00 GMT on 21 October; Google's blog post is dated 22 October. Google has not published prices or regions for the new models.
Disclosure: RECATOOLS is written with the assistance of Claude, made by Anthropic, whose Claude models are among those Google cites in this announcement. Readers should weigh that interest.
What Google and OpenAI announced
The Google post, written by Michael Gerstenhaber, vice-president of product management for Gemini at Google Cloud, announced the launch of OpenAI models in the model garden. It frames the move as a matter of choice. Customers, it says, can pick and swap models "based on performance, latency, and cost efficiency without architectural lock-in or separate vendor contracts."
OpenAI's post carries the headline "OpenAI models are coming to Google Cloud". The news-feed summary says the companies will put OpenAI's frontier models and Codex tools into Gemini's model garden, sold through Google Cloud Marketplace. The full OpenAI page refused our automated request, so this report relies on Google's text for detail.
Google also makes a competitive claim. It says it is now the only hyperscale provider offering access to all three leading closed-source model families, which the share text on its post names as Gemini, Claude and GPT. That is Google's own description of its position and has not been independently checked here.
Astra, Sol and Luna, by job
Google lists the three models by job. It aims GPT-6 Astra at the hardest work, among it coding, computer use and scientific research. Sol, in its 6.1 release, is pitched as close to Astra in intelligence for less money. GPT-6 Luna is the quick, cheaper option for routine work in volume.
Those descriptions come from Google and OpenAI. The post cites no benchmarks and no per-token rates for any of the three models.
Who can use them, and when
Access is limited. Google says the models are going to "select enterprise customers", and it asks others to fill in an expression-of-interest form. That form states plainly that submitting it "does not guarantee access."
Customers in the preview can call the models through Google's APIs. They can also point their local Codex command-line and desktop tools at Google's platform for inference. Google says the models will later appear in Gemini for business and in Google Antigravity, but gives no date.
The billing arrangement may matter more to finance teams than the models do. Because the OpenAI models are bought through Google Cloud Marketplace, usage appears on the customer's normal Google Cloud bill, Google says. It can be paid from cloud commitments already made and counts toward Google's Flexible Savings Plans, so a company with committed Google spend can use it on OpenAI models without a separate contract.
Security promises, two still pending
Google says customer prompts, context and outputs stay within a perimeter the customer sets, and that it does not use them to train foundation models. OpenAI's models are hosted on infrastructure Google dedicates to them, and queries pass through isolated gateways. Administrators get audit logging and VPC Service Controls protection.
Zero Data Retention options and Confidential Compute are "coming soon" rather than available today. Buyers with strict retention rules should check those dates before moving regulated work across.
The post carries one customer quote, from Yashodha Bhavnani, vice-president of AI products at Box. "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," she said. The future tense matters: Box is describing a plan, not a deployment.
The same models on AWS and Azure
Google is late to this. AWS announced on 8 September that GPT-6 Astra was generally available on Amazon Bedrock, with a context window of up to one million input tokens. Microsoft's Foundry model catalogue already lists gpt-6-astra, gpt-6-luna and gpt-6.1-sol.
Google's preview does not give developers a model they could not already reach. It changes where the bill goes. For an organisation that runs mostly on Google Cloud, the OpenAI models can now sit under the same contract, identity system and audit trail as Gemini.
What is still missing
Google's own documentation has not caught up. Its OpenAI models page, last updated on 22 October, still lists only the open-weight gpt-oss 120B and gpt-oss 20B, served from the global endpoint and us-central1 respectively. The platform's release notes had flagged the managed gpt-oss-20b endpoint for retirement on 21 October, alongside several other open-model endpoints.
No regional availability has been published for the GPT-6 models, including for Google's Singapore region, and there is no rate card. Teams that want to compare costs across the three clouds can model token volumes in our LLM cost calculator, but the Google figures will have to wait for Google.
The headline claim is also untested while the preview remains private. Until the models reach general availability, Google's claim to carry all three leading model families holds only for the customers it has let in.
View author profile → · Editorial policy
