Google Cloud is adding OpenAI’s GPT models to its platform, giving enterprise customers access to OpenAI, Google Gemini and Anthropic Claude through the same cloud provider. The rollout begins with a private preview for selected enterprise customers, with requests for early access now open.
The agreement covers three models: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. OpenAI models are also planned for Google’s Antigravity agent-based coding application and Gemini for Business, extending the arrangement beyond cloud API access. Google says the addition makes it the only major cloud provider offering models from all three companies.
The three OpenAI models target different workload requirements. Astra is positioned for demanding work such as software development and research. Sol offers a lower-cost alternative with similar performance, while Luna is intended for everyday tasks where speed and efficiency are priorities. Specific pricing and a date for general availability have not been provided in the announcement details.
The models will join Google Cloud’s Model Garden, which already contains more than 200 options. Its existing selection includes Gemini and Claude alongside open models such as Mistral, Llama and Google’s Gemma. The OpenAI additions are expected to appear there in the coming weeks, while the initial private-preview rollout is already underway.
For enterprise buyers, the arrangement combines model access with existing cloud procurement. OpenAI inference will run entirely on Google’s infrastructure, with transactions handled through Google Cloud Marketplace. Charges will appear on customers’ Google Cloud bills and count toward existing spending commitments, allowing organizations to use the models without setting up a separate billing relationship for that usage.
Google says customer data will not be used to train the models. That assurance addresses a common concern in enterprise AI deployments, where prompts and responses may contain proprietary code, internal documents or other sensitive business information. It is separate from the question of model capability, but can be an important consideration when businesses decide which services to approve.
Early-access customers will be able to reach the models through an API, including when working with Codex CLI or Codex Desktop. In those configurations, the inference will take place on Google’s systems. Google’s platform also supports moving workloads among models based on requirements such as performance, response time and efficiency, rather than requiring every task to use the same model.
The Antigravity integration will expand a coding environment that already supports Gemini and Claude. Google has not specified which OpenAI models will be included in that client, so the cloud lineup does not yet establish the precise selection developers will see inside the application.
The deal adds another distribution channel for OpenAI without replacing its principal cloud relationship: Microsoft remains its primary cloud partner. For Google, it broadens the models available to customers even as its own Gemini family competes for the same workloads. Access to the new cloud offering remains limited to approved private-preview participants.
Sources: The Decoder, Ars Technica (AI)