Google Cloud adds OpenAI models in private preview
GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are becoming available in Google Cloud's private preview. Users can access them through APIs and use them with Codex CLI and Desktop on their local computers.
Google Cloud is adding OpenAI's newest models to its Model Garden. This makes Google Cloud the only large-scale cloud provider to offer all three leading closed-source large language models (LLMs), along with more than 200 models made by Google, other companies, and open-model developers. This gives business teams one place on Google Cloud to choose models based on their abilities, speed, and cost. They do not need separate contracts with different vendors or major changes to their existing systems.
The first group includes three models for different needs. GPT-6 Astra is built for difficult coding tasks, using computers, professional work, and scientific research. GPT-6.1 Sol offers intelligence close to Astra but costs less. It is designed for AI agents that write code, use computers, and handle professional tasks. GPT-6 Luna is made for everyday tasks that need fast results at a low cost, even when used on a large scale. Customers can choose different models or switch between them for each task through Model Garden.
OpenAI models are coming to Google Cloud.
— OpenAI (@OpenAI) October 21, 2026
We're bringing our frontier models and Codex agentic workflows to Google Cloud's Model Garden, so enterprises can use our technology where they already build.https://t.co/gjuLVALjDm
Customers will pay for OpenAI models directly through Google Cloud Marketplace. The charges will appear on their usual Google Cloud bill, count toward existing cloud spending commitments, and qualify for Flexible Savings Plans that cover models from different providers. This setup is meant to give organizations more model choices while letting them manage purchases and costs through their existing cloud account.
Google is also applying its business security and management controls to these models. Prompts, context, and outputs will stay within the customer's chosen data boundary and will not be used to train foundation models. Options that prevent data from being stored, known as Zero Data Retention, are expected soon. The OpenAI models will run on dedicated Google Cloud systems through separate, isolated gateways. Google also plans to add Confidential Compute, which helps protect data while it is being processed. Identity checks, VPC Service Controls, shared audit logs, and tools for administrators to monitor activity will cover both models and AI agents. Google says outside model providers will not be allowed to access customers' cloud environments without permission.
GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are now becoming available in private preview to selected business customers. Those taking part can use the models through APIs and connect their local Codex CLI and Desktop workflows to Google's Gemini platform to run them. Businesses can request access by filling out an interest form, and more customers are expected to get access later. Google also plans to make OpenAI models available through Gemini for business and Google Antigravity. Box said this arrangement would let it combine OpenAI's most advanced models with Google Cloud's security and compliance controls when working with business content.
Sources and related context
- Google Cloud's flexible billing and AI cost controls: Google's earlier announcement about billing explains the Flexible Savings Plans and shared AI spending controls mentioned in the OpenAI rollout.
