OpenAI plans Google Cloud rollout as selected enterprises enter private preview

OpenAI announced on October 21 that its frontier models and Codex agentic workflows are coming to Google Cloud’s Model Garden. General availability is planned for the coming weeks. Google’s accompanying announcement says a private preview is rolling out to selected enterprise customers, separating limited initial access from the broader release OpenAI has scheduled. The plans pair model access with Google Cloud’s enterprise infrastructure.

Google lists GPT-Astra, GPT-6.1 Sol and GPT-6 Luna for the preview. Its positioning spans demanding coding, computer use and research with Astra, lower-cost professional and agentic work with Sol, and fast everyday tasks with Luna. The announcement provides no new comparative benchmark.

Purchases will run through Google Cloud Marketplace, and OpenAI says enterprises can apply them against existing Google Cloud spending commitments. Google adds that usage will appear on a single cloud invoice and count toward its Flexible Savings Plans across model providers. The arrangement makes procurement part of the integration: a company can acquire the models through its cloud purchasing channel rather than treating model access as a wholly separate commercial relationship.

Google says preview customers can use its APIs and connect local Codex CLI and Desktop workflows to its Gemini platform for inference. OpenAI also identifies Google Antigravity and the Gemini Agent as forthcoming destinations for its models. Availability in those products remains a future step. The announcement covers both access to underlying models and plans to bring them into software development and business workflows.

Security features span controls described for the integration and additions that remain pending. OpenAI highlights built-in access management and centralized data governance. Google says workloads run on dedicated Google Cloud infrastructure, with authenticated identities, unified audit logs and VPC Service Controls protection. It states that customer prompts, context and outputs are not used to train foundation models. Zero Data Retention options and Confidential Compute are described as coming soon, rather than capabilities already generally available.

Model Garden’s existing documentation describes a library for discovering, testing, customizing and deploying models from Google and partners. Its model cards provide the details for individual offerings, while the platform integrates with evaluation, tuning and serving tools. Administrators can restrict access to particular models through organization policies at the organization, folder or project level. Those documented controls provide context for the partnership’s promise of model choice: access can still be limited to models an enterprise has reviewed, instead of exposing every model in the catalog to every team.

Marketplace documentation separately sets out the purchasing layer. Enabling a model requires the appropriate Identity and Access Management permissions and acceptance of the applicable terms; some offerings also request information about intended use. Terms presented during purchase or deployment are the current versions. The documentation notes that some models have location-dependent pricing, determined by the endpoint receiving requests, and that eligible third-party models can draw against an organization’s Flexible Savings Plans. These are general Marketplace provisions, not a published price list or regional availability map for the newly announced OpenAI offering.

Microsoft remains OpenAI’s primary cloud partner, according to OpenAI’s announcement. The Google agreement adds a distribution and purchasing route while that relationship continues. Neither company’s announcement specifies per-token prices or a complete regional rollout schedule for this integration. For wider access, OpenAI is still pointing to general availability in the coming weeks, beyond the selected enterprises covered by Google’s private preview.