OpenAI on Google Cloud: Pricing & Access (Oct 2026)
OpenAI models enter Google Cloud private preview. Billing, cloud commitments and access explained; token rates unannounced.
By AI Pricing Guru Editorial Team
The AI Pricing Guru editorial team verifies prices against official provider pages and maintains the datasets, calculators and guides. See our methodology and contact us to report a correction.
Report summary — 2026-10-21
- Google says GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out in private preview to selected enterprises. OpenAI says general availability is planned in the coming weeks—not available to everyone today.
- Purchases will run through Google Cloud Marketplace and can draw down GCP spend commitments. Google also says usage counts toward Flexible Savings Plans; that is not a published token-price cut.
- Neither announcement gives Google Cloud token rates. Request access and written commercial terms before assuming direct-API price parity or migrating production.
Historical story text, scoped to the publication date above:
OpenAI models are coming to Google Cloud, with private-preview access starting October 21. The immediate change is an enterprise access and purchasing route—not a newly published inference discount.
OpenAI’s announcement confirms Model Garden distribution, Marketplace transactions and cloud-commitment drawdown. Google’s announcement names the initial models and preview restrictions.
What changes for access and purchasing
| Buying consideration | Direct OpenAI route | Announced Google Cloud route |
|---|---|---|
| Access | Separate OpenAI API account and terms | Private preview for selected enterprises; broader availability planned |
| Initial model lineup | Check OpenAI’s own catalog | GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna |
| Billing | OpenAI billing relationship | Google Cloud Marketplace; single Google Cloud invoice |
| Existing GCP commitments | Not the purchase route described here | Eligible purchases can draw down commitments |
| Commercial terms | OpenAI’s own rate card applies to its service | Token rates and any price difference absent from both announcements |
| Pricing data: How we verify → | ||
How we verify pricing and availability. Announcement comparison checked October 21, 2026; no Google Cloud rate card is inferred.
Google says selected customers can use its APIs and connect local Codex CLI and Desktop workflows to its Gemini platform for inference. That does not mean model weights run locally, or that an existing ChatGPT/Codex subscription includes Google Cloud inference.
OpenAI also says Microsoft remains its primary cloud partner. This announcement adds a distribution channel; it does not announce an exclusive move away from Microsoft.
Pricing impact: cloud budgets, not a confirmed discount
The biggest commercial change is where spending can count. Enterprises with unused GCP commitments may be able to purchase OpenAI inference against an already-contracted cloud budget. That can simplify procurement and improve commitment utilization without reducing the underlying token rate.
Google says usage also counts toward Flexible Savings Plans across model providers. Neither announcement quantifies an OpenAI-specific discount, publishes an input/output rate card, or establishes parity with OpenAI’s direct API.
Keep three numbers separate: the inference charge, any applicable contract discount, and the reduction in an outstanding cloud commitment. Commitment drawdown is not free inference, and counting toward a savings plan does not establish your final effective rate.
Use our OpenAI pricing page for direct-API reference rates and Google AI pricing page for Google’s own model pricing. Neither should be read as the new Google-hosted OpenAI rate card. Our API cost guide explains the other items a migration budget needs.
What enterprise teams gain—and must still check
Potential winners: companies already operating on Google Cloud, especially those with existing commitments, centralized procurement or a requirement to evaluate multiple model providers within one governance environment.
Who should wait: teams whose case depends on a lower token bill, immediate public access, or specific compliance features not yet delivered. No measured performance or cost advantage over direct access is established by these announcements.
Google describes dedicated Google Cloud infrastructure, identity controls, VPC Service Controls and unified auditing. It says prompts, context and outputs stay within the designated data perimeter and are not used to train foundation models. These are vendor statements, not our independent security audit.
Crucially, Zero Data Retention options and Confidential Compute are described as coming soon. Google Antigravity and Google’s broader Gemini business/agent integrations are also future availability—not features to assume are enabled today.
What to do now
- Request preview access. Use Google’s interest form or OpenAI’s access form. Ask which models, regions and quotas your organization actually receives.
- Get the complete quote. Confirm token rates, cache and batch terms, context-related charges, commitment eligibility, applicable savings-plan terms and support costs. Ask what is generally available versus preview-only.
- Run a matched pilot. Compare the same prompts, token volumes, retries, latency and failure handling against your current route. The token calculator can establish a direct-API baseline; use Google’s quoted rates separately until a verified rate card is available.
- Keep the existing route until the checks pass. Validate contractual data handling and required controls before sending sensitive workloads. A procurement advantage alone does not prove a production migration is ready.
Sources and verification
Both official announcements were checked October 21, 2026. This is an availability announcement, not a new model release or verified price cut. Google describes private preview now; OpenAI plans general availability in the coming weeks. Neither provides an exact general-availability date or Google Cloud token rates. We did not run paid inference or independently test the preview.