In this article
On October 21, 2026, Google Cloud announced that it now offers OpenAI's three models, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, in Model Garden, making it the only major cloud provider that puts Gemini, Claude and GPT on one platform, alongside more than 200 other models. What it means for your company: you pick the best model for each task from one supplier and pay on a single Google Cloud invoice that draws down your existing cloud commitments. But the OpenAI models are still in private preview for select enterprise customers, you request them through an interest form, and they are not yet listed in the Dammam region, where Claude and Gemini already are.
What exactly did Google Cloud announce?
According to Google Cloud's announcement, written by Michael Gerstenhaber, VP of Product Management for Gemini, OpenAI models are arriving in Model Garden on Gemini Enterprise Agent Platform, the platform Google describes as the evolution of Vertex AI since April 2026. These are the three models as the announcement describes them:
| Model | How Google describes it | Status |
|---|---|---|
| GPT-6 Astra | For demanding work: coding, computer use, professional tasks and scientific research | Private preview |
| GPT-6.1 Sol | Near-Astra intelligence at lower cost | Private preview |
| GPT-6 Luna | Fast and cost-efficient for everyday work at scale | Private preview |
Google says OpenAI models are also coming soon to Gemini for business and to the Google Antigravity coding tool, with no date given. For the differences and prices of the three models, read our article on GPT-6 Astra vs Sol vs Luna.
What changes for your company in practice?
- One invoice: you buy OpenAI models through Google Cloud Marketplace, and all usage appears on a single Google Cloud invoice.
- Your commitments count: usage draws down your existing cloud commitments and counts toward Flexible Savings Plans across model providers.
- One contract and account: no separate accounts with OpenAI, Anthropic and Google, just one Google Cloud project.
- Same developer tools: you reach the models through APIs, and you can connect Codex CLI and the Codex desktop app directly to the platform for inference.
- Unified oversight: authenticated identity for every agent and model interaction, unified audit logging, and VPC Service Controls protection.
One example in the announcement is Box, which uses OpenAI models through Google Cloud to give its customers frontier models together with Google Cloud's security and compliance controls.
What does Google say about your data?
The announcement lists three commitments, two of which include parts that have not arrived yet:
| Commitment | Available now | Coming |
|---|---|---|
| Data perimeter | Prompts, context and outputs stay within the designated data perimeter and are never used to train foundation models | Zero Data Retention options |
| Isolated running | Models run on dedicated Google Cloud infrastructure, and data moves only inside its environment | Confidential Compute |
| Auditing and identity | Authenticated identity, end-to-end visibility, unified audit logging | Nothing announced |
Do these models run from the Dammam region?
This matters for any Saudi company that cares where its data is processed. The official locations page lists Dammam (me-central2) among three Middle East regions, showing Gemini 4 Argon and other Gemini models, plus 17 Claude models, including Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1. GPT models do not appear in the Dammam list yet.
Note two points Google makes itself. First, endpoints do not guarantee data residency or in-region ML processing. Second, the data residency page lists the regions where Google commits to keep ML processing within the country's jurisdiction, and it does not mention Dammam or Saudi Arabia yet. If your contract or your sector's rules require data to be processed inside the Kingdom, get written confirmation from Google before you send it sensitive data.
How do you put it to work, step by step?
- Sort your tasks first: separate heavy work such as contract analysis or coding from high-volume daily work such as triaging messages and summarising, since each calls for a model at a different price.
- Start with what is available: Gemini and Claude are on the platform today, so run the same task on both and compare quality and cost.
- Register interest in OpenAI models: through the official interest form, and do not build a production system on a model still in private preview.
- Choose the region on purpose: use a regional rather than the global endpoint if processing location matters to you, and check the data residency page before launch.
- Make switching models easy: build your system so the model name comes from configuration, letting you move between Claude, GPT and Gemini without rewriting code.
- Watch the bill: one invoice is easier to follow but mixes three providers' usage, so set spending limits per project.
For a practical comparison of two low-cost models from different companies, read Claude Haiku 5.5 or GPT-6 Luna?, and for the idea of splitting one task across several models, read about Sakana's Fugu.
When do you not need this change?
If you use a single model directly from its provider and pay that provider, with no cloud commitment to Google, the announcement will not change much in your day. The biggest gain is for companies already running on Google Cloud, those that want to try several providers without several contracts, and those that need one audit trail for all their AI use.
Sources
Frequently asked questions
Are OpenAI models available on Google Cloud now?
Yes, since October 21, 2026 in Model Garden, but in private preview for select enterprise customers. You request access through the official interest form.
Which OpenAI models does Google Cloud offer?
Three: GPT-6 Astra for demanding work, GPT-6.1 Sol with near-Astra intelligence at lower cost, and the fast, low-cost GPT-6 Luna.
Do I pay OpenAI separately?
No. You buy through Google Cloud Marketplace, and usage appears on one Google Cloud invoice and draws down your existing cloud commitments.
Are GPT models available in the Dammam region?
Not yet. The official locations page lists Gemini and 17 Claude models in Dammam (me-central2), but no GPT models.
Does Google use my data to train models?
Google says prompts, context and outputs are never used to train foundation models, and that Zero Data Retention options are coming soon.
Does the Dammam region guarantee my data stays in Saudi Arabia?
Not automatically. Google says endpoints do not guarantee data residency, and its data residency page does not mention Saudi Arabia yet. Get written confirmation if you need it.
Follow Origami in Google
Pin Origami as a preferred source and our articles will surface first for you in Google Search and Top Stories.
