OpenAI and Google Cloud said on October 21 that GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna and Codex are now available in Google Cloud's Model Garden through a private preview, with general availability planned for the coming weeks. The listing means enterprise customers can purchase access to OpenAI's models through the Google Cloud Marketplace and draw down existing GCP spending commitments to pay for them. Both companies published the announcement at 12:00Z, and Google Cloud's own blog corroborated the terms, giving the claim support from both sides of the deal rather than from OpenAI alone.
The arrangement puts a direct OpenAI competitor's infrastructure in the position of reselling OpenAI's own models, which is notable given that Google develops its own Gemini line and competes with OpenAI for the same enterprise accounts. Model Garden has functioned as Google Cloud's aggregation point for third-party and open models, so adding OpenAI's flagship lineup extends that catalog rather than replacing anything already there. The models are also set to come to Antigravity and Gemini Agent, though neither company detailed what integration into those products will look like or when it will ship beyond the general preview-to-GA timeline.
What the announcement demonstrates is procurement flexibility, not technical parity or pricing parity with OpenAI's own API. Customers already holding GCP commitments gain a path to use OpenAI models without opening a separate billing relationship, which matters for finance and procurement teams more than for model selection itself. The companies did not disclose pricing, rate limits, regional availability, or whether Model Garden access includes the same feature set, context windows, or fine-tuning options available through OpenAI's direct API. Those gaps mean the listing should be read as a distribution and billing change first.
For researchers and operators evaluating deployment options, the practical items to check are whether Model Garden pricing tracks OpenAI's direct rates or carries a markup, whether latency and regional footprint match expectations for production workloads, and whether private preview access requires separate enrollment beyond standard GCP account status. Teams already budgeting against GCP commitments have a new lever to test Codex or GPT-6 Astra without renegotiating vendor contracts, but the preview label means behavior, availability, and terms could shift before GA. Anyone benchmarking now should treat results as provisional until the general availability release locks down the specifics.
The announcement did not specify how long the private preview phase will run or what criteria determine when GA begins, so the realistic next step is watching Google Cloud's own release notes rather than OpenAI's announcement for the operational details that matter to deployment planning.