Alphabet (NASDAQ: GOOGL)’s Google Cloud is rolling out selected OpenAI models in its Model Garden to a group of enterprise customers in private preview. The move puts model selection, commercial consumption and cloud controls on one platform, giving organizations a way to test frontier models without creating a separate procurement and operations path for each provider.
Google Cloud said GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are the first OpenAI models available through the program. The company characterizes Astra as a model for demanding coding, computer-use, professional and scientific work; Sol as a lower-cost option for agentic coding, computer use and professional work; and Luna as a faster, cost-efficient option for everyday work at scale. Customers can build with the models through Google Cloud APIs during the private-preview phase.
The practical enterprise issue is not simply access to another model. Model portfolios now vary by latency, price, reasoning behavior and task fit, while corporate buying, identity management and audit practices are still often organized around a small number of cloud platforms. Google Cloud said customers can dynamically choose and swap models for a use case based on performance, latency and cost efficiency instead of establishing separate vendor contracts for each option.
Google Cloud is also routing the commercial relationship through its Marketplace. The company said OpenAI model usage will appear on a single Google Cloud invoice, draw down existing cloud commitments and count toward Flexible Savings Plans across model providers. For a CIO or finance team, that arrangement can make model experimentation easier to attribute and govern than a collection of direct accounts with separate invoices and payment terms. It does not remove the need to compare actual application costs, but it gives that comparison a common billing surface.
The company is pairing the rollout with its existing platform controls. Google Cloud said customer prompts, context and outputs remain within a designated data perimeter and are not used to train foundation models; it said a zero-data-retention option is planned. It also said queries run through isolated gateways on dedicated Google Cloud infrastructure, while each model and agent interaction operates under an authenticated identity with unified audit logging and VPC Service Controls protection.
Those claims matter most when a model is connected to enterprise content or can take action through an agent. A customer still needs to define what data a model can reach, which tools it can invoke and which actions need approval. Bringing third-party model calls into the same identity and logging layer as the surrounding cloud application can give security teams a clearer control point, but it does not substitute for application-specific permissions and evaluation.
Google Cloud said additional OpenAI model availability is planned for Gemini for business and Google Antigravity. For now, the limited private preview makes this a staged adoption option rather than a broad production entitlement. Even so, it points to a significant market change from earlier AI architectures: enterprises increasingly expect to select among models while preserving the billing, identity and audit systems of their primary cloud provider, rather than treating every new model vendor as a separate operational island.