Google Cloud customers will be able to buy and run OpenAI models within Google's cloud platform. The announced route may simplify purchasing for businesses already building there, but it begins as a private preview for selected enterprise customers.
Google's launch details describe the preview. OpenAI's announcement says general availability is planned in the coming weeks. The arrangement does not establish access through an ordinary consumer account or change what a personal ChatGPT subscription includes.
Private preview first. Wider access is planned.
OpenAI models on Google Cloud, checked October 24, 2026.
- 01
Preview
Selected enterprise customers, with access approval.
- 02
Models
GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are listed.
- 03
Billing
Purchases through Google Cloud Marketplace.
- 04
Next
General availability and more product integrations are forthcoming.
Models and billing
The preview includes GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in Google's model garden. It also covers Codex workflows. Purchases are intended to run through Google Cloud Marketplace, with eligible spending drawing down existing cloud commitments.
Google describes broader availability in its business and developer tools as forthcoming. Access through one approved enterprise account therefore does not establish that a model is available in every Google app.
The route changes more than the visible model choice. Identity permissions, request handling, log access and the billing account all affect how an organization can use it. Marketplace eligibility and commitment treatment need to be confirmed for the intended organization and project.
Preview permission and data controls
OpenAI's access form identifies the preview as invite-only and intended for approved development and internal uses. Those terms do not by themselves authorize a wider production deployment.
Google says prompts, context and output in the preview are not used to train foundation models. Zero data retention and Confidential Compute are described as forthcoming. An organization that requires either capability needs confirmation that it has been delivered for the intended account before supplying the relevant data.
A successful internal evaluation would still leave production eligibility, retention terms and operating arrangements to resolve. For businesses with a working integration, the decision to migrate depends on whether the new purchasing and deployment route improves the actual workload enough to justify the configuration work.
What Marketplace purchasing changes
A company with an existing Google Cloud commitment is comparing two costs: the cash price of model usage and how that usage counts toward spending it has already promised. OpenAI's announcement says purchases can draw down GCP commitments. A commitment benefit can make the route attractive without establishing that the underlying model has a lower list price.
For example, a hypothetical $1,000 of eligible usage would reduce an outstanding commitment by $1,000 if it receives full credit under that customer's agreement. It would still represent $1,000 of model consumption. A company with no unused commitment receives a different financial benefit from the same transaction. Neither example establishes the actual preview price.
Google also describes a consolidated invoice. That can reduce the number of separate vendor bills an organization reconciles, while departments still need to allocate consumption to projects. A central purchasing route does not decide which team pays when an agent retries an unsuccessful task.
| Decision | Announced Google Cloud route | What remains to be established |
|---|---|---|
| Purchasing | Marketplace transaction and Google Cloud invoice | Account eligibility and quoted rates |
| Commitment use | Purchases can draw down GCP commitments | Terms of the customer's agreement |
| Access | Selected enterprise private preview | Approved uses and production permission |
| Data training | Google says prompts, context and outputs are excluded | Retention and feature-specific terms |
| Isolation | Dedicated infrastructure described | Delivery of forthcoming Confidential Compute |
Model choice still belongs to the application
Google positions Astra for demanding work, Sol for a lower-cost balance and Luna for everyday volume. Those are the supplier's product descriptions, not comparative results from Edit USA. A document-extraction service and a coding agent may justify different models even when the same purchasing department buys both.
For an existing application, moving providers can involve credentials, endpoint configuration, tool integrations and operational support. Include that engineering cost in the comparison with continued service through the present route. The preview is most compelling when the organization already wants Google's purchasing and governance arrangements. A consumer looking for extra ChatGPT access has a different product question; this enterprise access form supplies no such entitlement.
