The new OpenAI Google Cloud Model Garden listing is small for now: three models, one private preview and no launch date. On October 21, OpenAI and Google Cloud said that OpenAI's frontier models and its Codex coding agent will be offered through Google Cloud's Model Garden catalog. The move puts OpenAI on all three major US cloud platforms and changes how companies that already spend with Google can pay for OpenAI usage.
The rollout is narrower than the headline suggests. Reports describe the three launch models as being in private preview for select enterprise customers, not general availability. Those reports say general availability is planned "in the coming weeks," but no date has been set.
A note on sourcing: USA 24 could not retrieve the original OpenAI or Google Cloud announcements. The details here come from secondary write-ups, including cellcog.ai, Alphasignal and OpenTools. Read the official posts before making any purchasing decision.
Which OpenAI models reach Google Cloud first
The initial lineup has three models. Blockchain.news describes each by its role, and the table below summarizes those descriptions.
| Model | How reports describe it |
|---|---|
| GPT-6 Astra | OpenAI's highest-intelligence model |
| GPT-6.1 Sol | Balance of capability and cost |
| GPT-6 Luna | Efficient everyday model |
The labels help with budgeting decisions. Astra sits at the top tier by OpenAI's own description, while Luna is aimed at efficient everyday work. No benchmark or per-model price comparison appears in the material available for this report, so treat the labels as positioning rather than measured results.
Codex is also headed to Model Garden, with arrival expected within weeks, according to daily.dev. Whether Codex launches on the same timeline as the three models has not been confirmed. Codex is not new to rival clouds, either. One report says it reached AWS Bedrock by June.
Reports also say OpenAI models will appear in Google Antigravity and the Gemini Agent for task routing. USA 24 could not verify that detail independently.
How billing and data controls work on Model Garden
Customers buy access through Google Cloud Marketplace, and eligible Google Cloud spending commitments can be applied to the bill, according to Alphasignal. For a company that already has a committed-spend agreement with Google, that means OpenAI usage could draw down money it has already set aside. The cellcog.ai account of the announcement adds that usage counts toward commitments and Flexible Savings Plans.
Pricing is the largest unknown. No per-token rates were published, so there is no way yet to compare costs with OpenAI's direct API. Buyers should ask for per-model prices in writing before they route production workloads.
On data, Google Cloud has said the models will sit under its data-governance controls. Zero-data-retention options and confidential compute are planned for later, according to dev.to. Those features are not live yet. Any company with strict retention rules should confirm them before sending sensitive data.
Why Microsoft's role still complicates the story
Microsoft was OpenAI's exclusive cloud partner until recently, and that history is the backdrop for this deal. According to daily.dev, Microsoft's license with OpenAI became non-exclusive, which let other hyperscalers offer OpenAI's models and tools. One report ties the timing to an exclusive licensing deal that lapsed in April, followed by an expansion onto AWS, as described by Biggo Finance.
When one cloud controlled OpenAI's distribution, buyers had one main route in. Now the same models can be bought through several clouds, and the buying path depends on which contracts a company already holds.
The announcement also frames the relationship differently. As quoted by cellcog.ai, OpenAI's announcement says: "Microsoft remains OpenAI's primary cloud partner." That quote has no named speaker or title, so it reads as the company's framing rather than an individual's remarks.
Other coverage describes the Google deal as completing a sweep of the big three clouds. The accounts also disagree on order. Some place Google Cloud as OpenAI's third major US cloud after Azure and Oracle, while others tie the expansion to AWS. For context on OpenAI's finances, USA 24 has reported on the company's effort to close a $30 billion funding round before Anthropic's IPO.
What this means for enterprise buyers and Google
The strongest case for this move is in regulated sectors. Banks, healthcare providers and government agencies often buy through procurement and billing systems they already use. Buying OpenAI models through Google Cloud could let those buyers skip a new vendor setup, which may speed adoption, according to tech-insider.org.
For Google, the deal is a mixed result. Reports describe it as giving OpenAI sales access through a cloud run by a direct competitor, since Google develops Gemini. Google gains OpenAI demand inside its catalog, but it now hosts a rival's most advanced models beside its own. Model Garden previously showcased mostly Google's own models plus some open-weight options, according to tech-insider.org. Adding OpenAI's frontier models changes that mix in a visible way.
Do not mix this rollout with the early October launch of GPT-6 and the Sol and Luna models in ChatGPT. Those consumer rollouts came first, according to the LLM Gateway release timeline. The Google Cloud news is a separate distribution step aimed at businesses.
Steps for business buyers before routing workloads
Most readers have no direct action yet, because access is still in private preview. Enterprise buyers can take these steps now:
- Ask your Google Cloud account team whether your organization has preview access to the GPT-6 models.
- Get per-model pricing in writing, since no per-token rates have been published.
- Confirm data-retention terms, including whether zero-data-retention and confidential compute are live for your region and workload.
- Check whether your current Google Cloud commitments apply to OpenAI usage, and under what terms.
- Test on non-sensitive workloads before moving production traffic.
If you are not an enterprise customer, nothing changes for you today. You will not need to act until the models reach general availability.
Agent governance deserves attention too. Codex is an agentic tool, and USA 24 recently covered Microsoft's risk scores and kill switches for rogue AI agents. Any team that adds agents to a new cloud should set those controls before deployment.
What to watch as general availability approaches
Four open questions will decide how this lands. The first is the general availability date, and whether access will be limited by region or customer type, which the sources do not yet specify. The second is pricing, because a cost comparison with OpenAI's direct API is impossible until rates are published. The third is whether Codex arrives with the models or later. The fourth is whether the data-retention and confidential-compute features move from plan to product.
The Microsoft question sits above all four. If OpenAI keeps spreading its models across rival clouds, the phrase "primary partner" will matter less as a description and more as a contract term. Watch for the official posts to settle which framing is accurate. Watch for the first enterprise customers who say publicly what they pay, because that will show whether a cloud run by a rival can become a normal way for companies to buy OpenAI.
