OpenAI models on Google Cloud: availability, billing and governance

OpenAI models on Google Cloud are rolling out in private preview to select enterprise customers. GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are arriving in Model Garden, announced by OpenAI on 21 October and by Google Cloud on 22 October 2026.

Access starts with an interest form, and OpenAI says general availability is planned in the coming weeks. Zero Data Retention options and Confidential Compute are described as coming soon, not available today.

Sources checked on 26 October 2026. Advisories, prices and availability can change; confirm the linked primary sources before acting.

What is available today

Three models, through Google Cloud APIs, for customers admitted to the preview. Customers can also connect local Codex CLI and Desktop workflows to Google's Gemini platform for inference.

OpenAI models in Gemini for business and Google Antigravity are announced for later. Google also says it is now the only hyperscale cloud offering Gemini, Claude and GPT models together.

Billing through Google Cloud Marketplace

Usage appears on your Google Cloud invoice. Because the models are sold through Google Cloud Marketplace, spend draws down existing cloud commitments and counts toward Flexible Savings Plans.

The announcement did not publish per-token rates. Do not assume they match OpenAI's direct API prices; get a quote before you migrate a budgeted workload.

Direct access vs OpenAI models on Google Cloud

The models are the same family; the operating envelope differs. Compare these points before choosing.

Area	Direct OpenAI API	Google Cloud preview
Billing	OpenAI account	Google Cloud invoice, commitments, Flexible Savings Plans
Identity	OpenAI projects and API keys	Authenticated identity on every interaction
Audit	OpenAI platform logs	Unified Google Cloud audit logging
Data boundary	OpenAI data controls	Prompts and outputs stay in your designated perimeter; not used for training
Availability	Generally available	Private preview, select customers

Data handling and what is still coming

Google says prompts, context and outputs stay inside your designated data perimeter and are never used to train foundation models. Queries are routed through isolated gateways inside Google Cloud.

Zero Data Retention options and Confidential Compute are both marked as coming soon. If your data policy depends on either, wait for them to ship before moving regulated workloads.

Governance checklist for enterprise AI

Treat model choice as a routing decision, not a rewrite. These steps keep that option open.

• Send model calls through one internal gateway so a provider switch needs no application change.
• Record which models may process which data classes.
• Apply VPC Service Controls and review the unified audit logs Google describes.
• Keep an evaluation set so a model swap is measured, not assumed.

Our verdict: request preview access if Google Cloud is your main platform and commitments matter, but wait for general availability, ZDR and published pricing before moving regulated workloads. Our provider-neutral AI layer guide shows the routing pattern. For architecture help, see our SaaS development and AI automation services.

Sources and further reading

• Google Cloud brings OpenAI models to customers — Google Cloud
• OpenAI models are coming to Google Cloud — OpenAI

Image credits

Explanatory diagrams: Cloudsoft Zone, CC BY-SA 4.0. Featured photograph: Chad Davis, CC BY 2.0, via Wikimedia Commons. Google Datacenter - The Dalles, Oregon (17832143871): Tony Webster from Minneapolis, Minnesota, United States, CC BY 2.0, via Wikimedia Commons.