OpenAI to sell GPT-6 models through Google Cloud's Model Garden

Customers buy the models through Google Cloud Marketplace, so OpenAI usage appears on the same invoice as other Google Cloud services and counts toward existing spending commitments and Flexible Savings Plans. Google says the addition makes it the only hyperscale cloud offering all three leading closed model families: Gemini, Anthropic's Claude and OpenAI's GPT.

Three models, one shelf

The lineup splits three ways. GPT-6 Astra targets demanding coding, computer use, professional work and scientific research. GPT-6.1 Sol sits near Astra in intelligence at lower cost, aimed at agentic coding and professional work. GPT-6 Luna is the fast, cheap tier for everyday work at scale. Preview customers can call the models through Google's APIs and point local Codex CLI and Codex desktop workflows at Google's Gemini platform for inference. Customers can select or swap models for each use case through Model Garden, which Google says carries more than 200 first-party, third-party and open models. Anthropic's Claude had been available there for some time before the OpenAI deal.

Data stays inside the perimeter

Google says prompts, context and outputs stay inside the customer's designated data perimeter and are not used to train foundation models. The OpenAI models run on dedicated Google Cloud infrastructure through isolated gateways, and Google says third-party model providers do not get unauthorized access to customers' cloud environments. Zero Data Retention and Confidential Compute are planned for later phases rather than being available at launch. Box said the arrangement would let it pair OpenAI's frontier models with Google Cloud's security and compliance controls for enterprise content.

What the tokens cost

OpenAI's published rates put GPT-6.1 Sol at $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Cached input at $0.10 is one-twentieth of the $2 standard rate.

OpenAI has said GPT-6.1 Sol's Ultrafast tier costs $12 per million input tokens and $60 per million output tokens, which it described as 1.2 times the cost of Astra. That implies Astra at $10 per million input tokens and $50 per million output tokens, and puts Sol's output at one-fifth of Astra's.

Google has not published its own channel pricing, regions or a general availability date for the OpenAI models.

Why Google sells a rival's models

Thomas Kurian, chief executive of Google Cloud, said the company wanted to give customers a broad selection of models, including OpenAI's, so they could pick the right one for each job. Greg Brockman, OpenAI's president and co-founder, said the partnership made its most capable models and Codex available where customers already build. OpenAI said Microsoft remains its primary cloud partner.

Cloud distribution carries real revenue. Anthropic made 47% of its 2025 revenue through Amazon's and Google's cloud platforms, according to its IPO filing, Reuters reported. The OpenAI arrangement is not exclusive in either direction: AWS began hosting OpenAI's models in April after Microsoft's exclusive licence ended, and OpenAI added its agents to Amazon Bedrock last month. Mike Leone of Moor Insights & Strategy told Axios that Gemini now has to win those customers with OpenAI sitting right next to it.

What happens next

General availability is planned "in the coming weeks." Google also plans to bring the OpenAI models to Gemini for business and to Antigravity, its agent-powered coding tool. Preview access runs through an interest form, with more customers expected later.