OpenAI’s GPT-6 models reach Google Cloud, with no prices yet

OpenAI said in a post on X that its frontier models and Codex agentic workflows are coming to Google Cloud’s Model Garden, so enterprises can use the technology “where they already build.” Access starts as a limited preview.

A private preview with no price list

OpenAI’s post is short and gives no dates or prices. Fuller details come from secondary coverage, not from the post itself, and the company’s own announcement page could not be opened for this article.

According to TestingCatalog, three models are involved: GPT-6 Astra, aimed at demanding coding, computer-use, professional and scientific research work; GPT-6.1 Sol, described as close to Astra in intelligence at a lower cost; and GPT-6 Luna, built for fast, cost-efficient everyday workloads at scale. The report says the rollout is a “private preview for select enterprise customers,” with access requested through an interest form and more customers expected later.

Billing runs through Google Cloud Marketplace and appears on the same Google Cloud invoice, so usage can count toward an existing cloud commitment, the report says. Customers can call the models through APIs and connect local Codex CLI and Desktop workflows for inference. According to the same report, prompts, context and outputs stay inside the customer’s data perimeter and are not used to train foundation models. Zero Data Retention options are listed as coming soon, and Confidential Compute as planned.

Google is also expected to bring OpenAI models to its Gemini for business product and to Google Antigravity, per that report.

What is still missing for buyers

Per-token prices were not published by either company, according to Forkast, so Sol’s and Luna’s cost advantages cannot yet be measured in dollars. TestingCatalog notes that regions, context window sizes and benchmark figures are not given either. General availability is described only as expected in the weeks ahead.

One billing detail is disputed. Forkast says Google’s documentation gives third-party products no discounts from Flexible Savings Plans, and warns against assuming savings on these models, while TestingCatalog says charges count toward those plans. Buyers should confirm the terms with Google before planning around either reading.

The third hyperscaler in six months

The deal fits a pattern of OpenAI spreading beyond a single cloud. Forkast reports that OpenAI’s exclusive distribution through Microsoft Azure ended in April 2026, that Amazon Bedrock followed on April 28, and that Oracle Cloud came on June 10. Google Cloud, announced October 21, is the third hyperscaler to sell the models.

Google is not new territory for OpenAI technology: Forkast notes that the open-weight gpt-oss 120B and 20B models were already available on Google’s platform. The GPT-6 family is the closed tier, available only through APIs, so this is the first time Google customers can buy it from within Model Garden.

For Google, the move adds a rival lab’s models to a catalog it markets as a neutral shelf. For OpenAI, it puts its agentic coding tools in front of companies that standardized on Google Cloud, and lets them spend existing cloud budgets rather than open a separate contract. Reliability may also matter: OpenAI’s own service recently suffered a wide outage across ChatGPT, Codex and the API, and a second cloud route is one way enterprises can reduce dependence on a single provider, though neither company has said that is the aim.

The claims about model quality and data handling come from OpenAI-linked descriptions and have not been independently verified. Until Google and OpenAI publish prices, regions and a general availability date, the practical answer for most companies is to join the interest list and wait.