Google Cloud to offer OpenAI models through Model Garden
Google Cloud will add OpenAI’s newest AI models to its Model Garden platform. This will let business customers use these models through the Google Cloud systems they already have, including their security settings, management tools and business agreements.
The two companies announced the deal on October 21. The first models available will be GPT-Astra, GPT-6.1 Sol and GPT-6 Luna. According to Google Cloud’s announcement, a small group of selected business customers is starting to get early access to these models through a private preview. The companies plan to make the models available to more customers in the coming weeks.
Customers will be able to pay for using the models through Google Cloud Marketplace. They can also count these costs toward money they have already agreed to spend on Google Cloud services. The new OpenAI models will join more than 200 AI models already available on Model Garden. These include models made by Google, other companies and developers who share their models openly.
How businesses can use the models and keep control
The deal is meant to help companies building AI agents and other AI systems for everyday business use. AI agents are programs that can carry out tasks using AI. Companies need to compare different models to see what they can do, how much they cost and how quickly they answer. Google Cloud says customers will be able to choose between different models without having to make a separate purchasing agreement with each company that provides them.
Google Cloud also plans to support Codex tools that people use on their own computers, including the command-line tool and desktop app, through its cloud platform. The companies said more connections with Gemini for business and Google Antigravity are also planned for the future.
Google Cloud said its existing security and management tools will also work with OpenAI models used through the service. These include controls over who can access the models, a central record of activity for security checks and limits on how data moves across networks. Google Cloud said the questions and instructions customers send to the models, along with the answers they receive, will stay inside approved data boundaries. It also said this information will not be used to train the main AI models. Two other security features, Zero Data Retention and Confidential Compute, are planned for later. They will not be generally available when the service first launches. Zero Data Retention is meant to limit the storage of customer data, while Confidential Compute is designed to help protect data while computers are processing it.
These differences matter for companies deciding whether the early version of the service meets their security rules and legal requirements. Businesses will still need to check which regions offer the service, how private or sensitive information is handled and exactly which protections are included in their agreements.
More competition in cloud AI
By offering OpenAI models alongside its own Gemini models and models from other providers, Google Cloud wants Model Garden to become one place where businesses can manage different kinds of AI work. In this announcement, the companies did not say how many customers have started using the service. They also did not give exact dates for when it will become available in different regions or explain how much each model will cost.
The partnership gives businesses another way to use OpenAI models. It comes at a time when companies buying technology are thinking carefully about how much AI agents cost to run, how easily they can switch between different AI models and how difficult it can be to control AI systems that automatically access company tools and information.
For more details about how the services will work together, read the companies’ OpenAI announcement.