Companies using the Google cloud will no longer have to choose a single artificial intelligence provider for all their tasks. On October 21, 2026, Google Cloud announced the addition of three OpenAI models to its Model Garden catalog: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna.
With this addition, the company claims to be the only large-scale cloud provider to bring together the three major closed-source language models on the market, alongside more than 200 proprietary, third-party, and open models available on the same enterprise platform.
This is a commercial announcement from Google Cloud. For now, the new models are only available in private preview for selected customers, not in general availability.
Three OpenAI models for three types of work
The catalog does not incorporate a single model, but rather three tiers with different prices and capabilities, according to Google's own description.
GPT-6 Astra is the top tier, designed for demanding tasks such as advanced programming, computer use, professional work, and scientific research.
GPT-6.1 Sol offers, according to the company, a level of intelligence close to that of Astra but at a lower cost, with a focus on agent-based programming, computer use, and professional work.
GPT-6 Luna is the fastest and most economical model, oriented toward large-scale daily work.
Switching models without rebuilding architecture
Michael Gerstenhaber, Vice President of Gemini Product Management at Google Cloud, signed the announcement. He argues that modern enterprise artificial intelligence requires assigning each task to the most suitable model, rather than relying on just one for everything.
The technical promise is that customers will be able to select and switch models dynamically based on performance, latency, and cost. Google ensures that this change is made without architectural lock-in or the need for separate contracts with each provider.
A practical example would be a customer service team that sends simple queries to a lightweight model like Luna and reserves a superior model, such as Astra, for more complex cases. This distribution allows for cost containment without modifying the application.
A single invoice and leveraging spending commitments
The aspect that most interests purchasing departments is contractual. OpenAI models are contracted directly through Google Cloud Marketplace.
All consumption is reflected on a single Google Cloud invoice. Furthermore, that usage counts toward the spending commitments the company already has signed with the cloud and counts toward the Flexible Savings Plans of all model providers.
In practice, an organization with an already approved spending commitment with Google Cloud can absorb the consumption of OpenAI models within that budget, without the need to open a new line item with another provider.
What Google promises regarding data governance
Google maintains that running OpenAI models on its platform meets the same enterprise standards as the rest of the Gemini ecosystem. There are three specific commitments.
First, prompts, context, and responses remain within the data perimeter chosen by the customer and are not used to train foundational models. Zero Data Retention (ZDR) options will arrive soon, but are not yet available.
Second, OpenAI models run on dedicated Google Cloud infrastructure. The Confidential Compute feature is also announced as an upcoming addition. Queries pass through isolated gateways, and external model providers do not have unauthorized access to the customer environment.
Third, each agent and each interaction with a model operates under an authenticated identity. IT administrators have end-to-end visibility, protection via VPC Service Controls, and unified audit logging.
It should be noted that both ZDR and Confidential Compute are still listed as future features, not available at this time.
Box, the first customer to speak out
Content management company Box appears as a testimonial in the announcement. Yashodha Bhavnani, Vice President of AI Products at Box, stated that helping customers apply AI to corporate content requires leading models backed by robust security and rigorous compliance standards.
Bhavnani noted that by using OpenAI models through Google Cloud, Box will be able to offer its customers access to these frontier models alongside the security and compliance capabilities of Google Cloud. This is a statement of intent, not a deployment already in operation.
A new balance between clouds and AI labs
The move alters the competitive logic between cloud providers. Until now, each platform tended to partner with its own reference model. Google, which develops Gemini, is now also marketing the product of one of its main direct competitors.
The same week as the announcement, the Google Cloud blog reported that the Claude Opus 5.5 and Claude Sonnet 5.5 models, developed by Anthropic, are already available on Google Antigravity. The strategy seems consistent: to position the platform as a neutral ground where all the major models on the market coexist.
For AI labs, the cloud channel provides an additional path for enterprise distribution. For Google, it is a way to retain customer spending even if they do not choose Gemini for their tasks.
What changes if you already work with Google Cloud
Several practical points for technical teams emerge from the announcement.
Those who use Codex can now connect their local Codex CLI and Desktop flows to the Gemini platform for inference, according to Google. That connection is already operational.
To access the new OpenAI models, for now, only the private preview is available. It is necessary to request it through the interest form that Google has enabled, as there is no general availability yet.
What to review before taking the step
If your company has strict data retention or confidential computing requirements, it is advisable to wait for confirmation on the arrival date of ZDR and Confidential Compute before migrating sensitive workloads.
If you manage technology budgets, it is worth checking if the current spending commitment with Google Cloud can absorb the consumption of the new OpenAI models without the need for additional negotiations.
What it implies for the coming months
If your organization is already a Google Cloud customer, you will be able to test GPT-6 Astra, Sol, and Luna without signing an additional contract or managing a separate invoice, provided Google accepts your request in the private preview.
According to the announcement, the next step on the calendar is the arrival of OpenAI models to Gemini for business and Google Antigravity, although still without a specific date. Until then, the practical recommendation is to measure in real tasks which model performs best for every euro invested, as the platform promises precisely the possibility of switching from one to another without rebuilding anything.
Frequently Asked Questions
Which OpenAI models has Google Cloud added?
GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, with different levels of capability and cost, available for now only in private preview.
Can I use these models in my company yet?
Only if Google approves your request through the interest form for the private preview. There is no general availability yet.
How is the use of OpenAI models billed on Google Cloud?
All consumption appears on a single Google Cloud invoice and can be deducted from the spending commitments already signed with the platform.
Does Google Cloud also offer Anthropic models?
Yes. The same week as the announcement, Google confirmed that Claude Opus 5.5 and Claude Sonnet 5.5 are available on Google Antigravity.
Is data privacy guaranteed when using GPT-6 on Google Cloud?
Google states that prompts and responses remain within the customer's data perimeter, but key features such as Zero Data Retention (ZDR) and Confidential Compute are not yet available, only announced as upcoming.
Which company has confirmed the use of these models?
Box, a content management company, commented on the announcement through its Vice President of AI Products, Yashodha Bhavnani, although without confirming a deployment already in production.
