OpenAI models enter Google Cloud private preview
Selected enterprises get initial access; wider availability remains ahead.
Google’s announcement on October 21 names GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. Selected enterprises can use APIs and connect local Codex workflows. Google says customer inputs and outputs will not train foundation models. However, zero-retention options remain forthcoming features. These are different promises: excluding training does not itself mean that no data is retained.
OpenAI describes the partnership as a way to bring models and Codex’s software agents into the Google environment where developers already work. Its account pairs model access with Google’s existing access controls and centralized data governance. Purchases through Model Garden, the model catalog, go through Google Cloud Marketplace and can count against existing cloud spending commitments. That makes the procurement route relevant to teams already buying Google infrastructure. OpenAI plans general availability in the coming weeks, without naming a precise date. Microsoft remains its primary cloud partner.
For interested teams, the access form asks for business contact details, a company name, job title, country and industry. It explicitly says submitting interest does not guarantee access. Treat completing it as an application for consideration, rather than confirmation that your organization can start using the API.
The original report cites Google Cloud; it is a secondary account of the announcement.