OpenAI's GPT-6 Models Land on Google Cloud: What the Model Garden Deal Means for Microsoft and Brands

On October 21, 2026, Google Cloud said OpenAI's GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are coming to Google Cloud's Model Garden in private preview for select enterprise customers. Usage is billed through Google Cloud Marketplace and counts toward existing commitments. It is OpenAI's clearest step beyond Microsoft, and it puts GPT, Gemini and Claude on one hyperscaler for the first time.

What did Google Cloud and OpenAI announce?

Google Cloud is adding OpenAI's newest models to Model Garden, its catalog of more than 200 first-party, third-party and open models. The official Google Cloud blog post, written by Michael Gerstenhaber, VP of Product Management for Gemini at Google Cloud, describes three tiers:

• GPT-6 Astra: built for demanding coding, computer use, professional work and scientific research.
• GPT-6.1 Sol: close to Astra in capability at a lower cost, aimed at agentic coding and professional tasks.
• GPT-6 Luna: a fast, cheaper model for everyday work at scale.

OpenAI published its own announcement the same day. It speaks more generally about frontier models and Codex, says general availability is planned in the coming weeks, and points customers to an access form. Google Cloud CEO Thomas Kurian framed the deal as helping customers "choose the right model for any given job."

One small discrepancy: parts of Google's post call the top model "GPT-Astra," while OpenAI's own product pages and most coverage, including TestingCatalog, use "GPT-6 Astra."

Who can use it, and how does billing work?

For now, only select enterprise customers in a private preview can use the models, through Google Cloud APIs. Others can register interest. The commercial terms are the part enterprise buyers will notice first:

1. OpenAI usage is transacted directly through Google Cloud Marketplace.
2. It appears on the same Google Cloud invoice as everything else.
3. It draws down existing Google Cloud spend commitments and counts toward Flexible Savings Plans across model providers.

That last point matters: until now, using GPT models meant a separate contract with OpenAI or Microsoft. Now the same budget line can cover Gemini, Claude and GPT.

What does Google promise on data and security?

Google says prompts, context and outputs stay within the customer's designated data perimeter and are never used to train foundation models. OpenAI models run on dedicated Google Cloud infrastructure behind isolated gateways, with authenticated identities, VPC Service Controls and unified audit logs.

Two important controls are not live yet:

• Zero Data Retention (ZDR) options are listed as "coming soon."
• Confidential Compute is also still on the roadmap.

Techzine noted that these features seem to need more work, a fair point for regulated industries that will want ZDR before moving sensitive workloads.

What about Codex, Gemini for business and Antigravity?

Codex is the piece that works today. Starting October 21, customers can connect local Codex CLI and Codex Desktop workflows to Google's Gemini platform for inference, so coding agents can run under Google Cloud governance. Google also says OpenAI models will come to Gemini for business and its Antigravity agentic IDE, but gives no date. OpenAI's post separately mentions the Gemini Agent.

Why is this a big deal for Microsoft?

It is the most visible result of OpenAI loosening its Azure ties. According to Redmond Magazine's summary of the restructured deal announced in late April 2026:

• OpenAI can now serve its products to customers on any cloud.
• Microsoft keeps a license to OpenAI's models through 2032, but it is no longer exclusive.
• OpenAI products still ship first on Azure unless Microsoft cannot or chooses not to support them.
• The restructuring followed OpenAI's expanded partnership with Amazon earlier in the year.

OpenAI's announcement states that Microsoft remains its primary cloud partner. Google, for its part, still keeps Gemini exclusive to its own cloud, so the arrangement only flows one way. With this launch, Google can claim to be the only hyperscaler offering all three leading closed model families on one platform.

What are critics and skeptics saying?

The reaction has been more measured than celebratory. The main objections:

• It is still a preview. Only select customers have access, and key privacy features are not shipped. Skeptics on developer forums see it as another announcement ahead of real availability.
• Billing, not pricing. A dev.to analysis argues the change is about "billing convenience, not the underlying token economics." No price cuts were announced.
• Trust between rivals. Some commenters worry a direct competitor hosting OpenAI's models creates distillation or data risks, though others point out that contracts and isolated infrastructure make that unlikely.
• Concentration risk for Google. Crypto Briefing, citing UBS estimates, reports that OpenAI and Anthropic could make up more than 48% of Google Cloud revenue by 2027. Neither company is profitable, which gives them leverage on pricing and leaves Google exposed if they slow down.

What does this mean for marketers and brands?

The short version: the "one model" era is over even inside a single cloud. Gerstenhaber, quoted by Techzine, said workloads should be "matched to the right model, rather than relying on a single model." When enterprises can swap GPT, Gemini and Claude with one invoice, the apps, agents and assistants your customers use will run on a mix of models, and that mix will keep changing.

For brands, that has three practical effects:

1. Your brand is described differently by each model. GPT-6 Luna powering a retailer's support bot may recommend different products than Gemini in a search experience or Claude in an internal research agent.
2. Visibility in one assistant is not visibility everywhere. Tracking only ChatGPT misses the GPT-powered agents built on Google Cloud, and vice versa.
3. Content needs to work for all of them. Clear, citable, well-structured pages are still the best bet across models.

This is the problem MarkGrid's Model Share is built for: measuring how often and how favorably your brand appears across multiple AI models, side by side. If you are setting up a measurement plan, our guide to AI visibility metrics worth tracking is a good starting point, and the GEO guide covers how to shape content so generative engines cite it.

FAQ

Which OpenAI models are available on Google Cloud?

GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, in private preview through Model Garden. Google says more models are coming.

Can any Google Cloud customer use GPT-6 models today?

No. Access is limited to select enterprise customers in a private preview. Others can request access through an interest form, and OpenAI says general availability is planned in the coming weeks.

Does OpenAI usage on Google Cloud count toward my commitment?

Yes. Usage is billed through Google Cloud Marketplace, appears on your Google Cloud invoice, and draws down existing commitments and Flexible Savings Plans.

Is my data used to train models?

Google says prompts, context and outputs stay within your data perimeter and are not used to train foundation models. Zero Data Retention and Confidential Compute are not available yet.

Does this end OpenAI's partnership with Microsoft?

No. OpenAI says Microsoft remains its primary cloud partner, and new OpenAI products still ship first on Azure under the April 2026 terms. The exclusivity, though, is gone.

Why should marketers care about a cloud deal?

Because more AI products will be built on a mix of models. How your brand shows up can differ by model, so measuring visibility across GPT, Gemini and Claude matters more than tracking any single assistant.

Last updated October 22, 2026.