OpenAI will bring its models to Google Cloud, reaching all three major U.S. clouds

Google Cloud opens private access to OpenAI models for enterprise customers

Select enterprise customers can now use OpenAI models through Google Cloud and put the usage on their Google Cloud bill. Google’s October 21, 2026 announcement starts a private preview of GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in Model Garden—not a new OpenAI model release or a rollout open to everyone.

The change is where businesses can access the models. OpenAI introduced Astra in September, offering it through its own API, Microsoft Azure and AWS Bedrock. Its announcement page later added updates for Luna and Sol. Google Cloud is adding the models to its catalog, which includes more than 200 first-party, third-party and open models.

Customers admitted to the preview can build through Google Cloud’s APIs—the interfaces applications use to send requests to a model. They can also connect local Codex command-line and Desktop workflows to Google’s Gemini platform for inference, the computing that produces a model’s response. That gives those workflows a Google Cloud path for running the OpenAI models.

The commercial route runs through Google Cloud Marketplace. Google says OpenAI usage appears on one Google Cloud invoice, draws down existing cloud commitments and counts toward Flexible Savings Plans across model providers. The arrangement therefore covers both access to the models and how their usage fits into a customer’s existing Google Cloud spending.

Google also says teams can select and swap models according to performance, response time and cost without separate vendor contracts. Its stated approach is to match different workloads to different models, rather than make one model handle every task. Model Garden supplies the catalog; the unified platform supplies the access and purchasing framework.

Google positions the three options by workload and cost. These are the company’s descriptions of their intended roles, including its comparative claim about Sol’s intelligence, rather than independent findings about performance:

GPT-6 Astra is intended for demanding coding, computer use, professional tasks and scientific research.

GPT-6.1 Sol promises near-Astra intelligence at lower cost for coding agents, computer use and professional work.

GPT-6 Luna is positioned as a fast, cost-efficient option for everyday work at scale.

Google says customer prompts, context and outputs stay within the customer’s designated data perimeter and are not used to train foundation models. Zero Data Retention is a separate, forthcoming option—not a feature Google describes as available at the start of this preview. The announcement labels it as coming soon.

Google says the models run on dedicated Google Cloud infrastructure. Queries pass through hardened, isolated gateways, and data in transit stays within its cloud environment. Confidential Compute is listed as coming soon. Google also promises that third-party model providers cannot access a customer’s cloud environment without authorization.

For administrators, Google says each agent and model interaction runs under an authenticated identity. It also provides unified audit logging and VPC Service Controls, its controls for protecting service boundaries. Those features address oversight of who is interacting with models and the record of that activity across the enterprise.

The current access route should not be confused with every planned integration. Google says OpenAI models will also come to Gemini for business and Google Antigravity. Both remain future destinations in this announcement; the rollout starting now is the Model Garden private preview for selected enterprise customers.