On 21 October 2026, Google Cloud announced that OpenAI's GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out in its model garden, in private preview "to select enterprise customers." Our read: OpenAI models on Google Cloud are a procurement option to register interest in, not something to migrate production onto. The interesting part is the billing route, not the models. For the wider view of how releases like this change budgets, see AI model releases and pricing: what changes for engineering teams.
What Google announced, and what it didn't
Google Cloud's 21 October announcement says Google is "launching OpenAI models in Google Cloud's model garden." The three models are in private preview, access is by interest form, and more models are promised "soon." Codex CLI and Codex Desktop can connect to Google's platform for inference. OpenAI models are also due "soon" in Gemini for business and Google Antigravity.
Google also claims it is now the only hyperscale cloud offering access to all three leading closed-source LLMs. That's marketing, so treat it as Google's claim.
What the post leaves out matters more for planning. It gives no regions, no prices, no general availability date and no model IDs. Google's own documentation page for OpenAI models in its platform still lists only the gpt-oss models, with no GPT-6 entry as of 22 October. That page also says OpenAI models are not a Google product and sit under separate terms and each model's card. The docs haven't caught up with the announcement, which tells you how early this is.
For list prices, use our write-up of GPT-6.1 Sol pricing against Astra. Google's post states none for its own route, and we won't guess.
How the money moves: Marketplace, invoice, commitment, FSP
Here's the sentence that makes this a finance story. Google says that because OpenAI models are transacted directly through Google Cloud Marketplace, "all usage appears on a single Google Cloud invoice, draws down existing cloud commitments, and counts toward Flexible Savings Plans (FSP) across model providers."
Read that against Google's Flexible Savings Plan documentation. An FSP is a spend-based committed use discount: you commit to a monthly spend for one or three years. A one-year plan gives a 10% discount and a three-year plan gives 20%, on eligible Gemini Enterprise SKUs. Then the catch, in Google's words: "Third-party products don't get any discounts from the FSP." Third-party spend draws down your enterprise commitment, "subject to the standard Cloud Marketplace drawdown rules and contract caps defined in your enterprise agreement."
Counting toward an FSP is drawdown of the commitment, not a discount, per the FSP doc. Our judgment: the win here is one invoice and burning down committed spend you'd otherwise owe anyway. It isn't a cheaper token. The contract cap in your enterprise agreement decides how much of your OpenAI usage counts toward the commitment, and Google's post doesn't say what that cap is for you. Get the number in writing.
Google's FSP page says this for third-party models generally. It doesn't mention OpenAI by name, so confirm the OpenAI models fall under the same rule.
Google Cloud vs a direct OpenAI contract vs Azure vs Bedrock
| Route | How it's billed | Commitment drawdown | Status |
|---|---|---|---|
| Google Cloud | Marketplace transaction, single Google invoice | Yes, per Google, subject to contract caps | Private preview, select customers |
| AWS Bedrock | AWS | Per AWS, OpenAI models and Codex usage "can be applied toward existing AWS cloud commitments" | Limited preview at the April 2026 announcement; Managed Agents in public preview since 29 September |
| Azure | Microsoft says Azure-sold models are billed through your Azure subscription | Ask Microsoft | Microsoft lists all Azure OpenAI models among models sold by Azure |
| Direct OpenAI | Separate vendor contract and invoice; terms not covered here | Check your contract | Check your contract |
On AWS, our write-up of Bedrock Managed Agents with OpenAI covers the runtime limits, which moved to public preview on 29 September in three US regions. That page is about agent features. This one is about contracts.
We found no Microsoft statement on how Azure OpenAI spend counts against a Microsoft commitment, so we don't claim one. Ask. The same goes for your direct OpenAI contract: its terms are between you and OpenAI, and we haven't reviewed them.
Our decision rule, and it's judgment: carry the spend on the cloud where you already have an unburnt commitment and where the feature you need is generally available. Google's route has no GA date yet.
What "private preview to select enterprise customers" means for your timeline
You express interest through Google's form, and Google selects customers. The post gives no general availability date and no service-level statement. Assume nothing before GA.
Practically: don't put OpenAI-on-Google on a Q4 roadmap, and don't size a new Google commitment around expected OpenAI usage. If you're renewing, the safe version is a commitment you could burn down without OpenAI at all, with this as upside.
What is still "coming soon": ZDR and Confidential Compute
Google's data-handling bullets mix present and future tense, and the future-tense ones are the controls regulated buyers ask about first. The post says "Zero Data Retention (ZDR) options available coming soon," and that OpenAI models run on dedicated Google Cloud infrastructure "with Confidential Compute coming soon."
What's stated in the present tense: prompts, context and outputs "remain entirely within your designated data perimeter and are never used to train foundation models." Google also lists VPC Service Controls and unified audit logging.
For EU, health or other regulated workloads, wait until ZDR and Confidential Compute ship and a region list exists. The post names no region, so residency can't be assessed today. If you're comparing against the direct route, our piece on the OpenAI API BAA going self-serve lists what that contract covers and what you still owe.
What this means for your AI budget
Don't move an OpenAI forecast onto your Google commitment yet. Three questions for your Google account team, before any forecast changes:
- What is the Marketplace drawdown cap in our enterprise agreement, in writing?
- Are we on the preview list, and when would we get access?
- When will ZDR, Confidential Compute and a region list be committed in writing?
If your CFO owns a Google commitment with slack in it, the answers decide whether OpenAI spend can absorb that slack. If you're already committed on AWS or Azure, put the same cap question to that account team. AWS says OpenAI usage can be applied toward existing commitments, and neither AWS nor Microsoft states a cap.
To size what you'd be moving in the first place, see our breakdown of what AI coding tools cost per developer compared with hiring one.
FAQ
Can OpenAI models be used on Google Cloud?
Yes, in private preview and by request. Google announced on 21 October 2026 that OpenAI models are rolling out to select enterprise customers in its model garden. You express interest through a form, and Google picks who gets access.
Which OpenAI models are on Google Cloud?
Per Google's announcement: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. Google's documentation page for OpenAI models still lists only the gpt-oss models, so check access in your own project before planning around any of them.
Does OpenAI usage on Google Cloud count against my Google commitment?
Per Google, yes: usage is transacted through Cloud Marketplace and draws down existing commitments. It's subject to the standard Marketplace drawdown rules and the contract caps in your enterprise agreement, so ask your account team what the cap is.
Is there a Flexible Savings Plan discount on OpenAI models?
No. Google's FSP documentation says third-party products don't get any discounts from the FSP. The 10% (one-year) and 20% (three-year) discounts apply to eligible Gemini Enterprise SKUs. Third-party spend can still draw down your commitment.
Is zero data retention available?
Not yet. Google lists ZDR options as coming soon, along with Confidential Compute.
Sources
- Google Cloud: Google Cloud expands model choice and flexibility, bringing OpenAI models to customers
- Google Cloud docs: Flexible Savings Plans
- Google Cloud docs: OpenAI models in the Gemini Enterprise Agent Platform
- AWS What's New: Amazon Bedrock OpenAI models, Codex and Managed Agents (April 2026)
- Microsoft Learn: Foundry models sold directly by Azure