Google Cloud Opens to Rival OpenAI Models: GPT-6 Arrives in Model Garden

For years, the biggest names in AI have competed for the same enterprise budgets while keeping their flagship models largely on separate turf. On October 21, 2026, Google Cloud crossed one of the most visible lines in that rivalry: it announced that OpenAI models are now available in Google Cloud's Model Garden, starting with a private preview for selected enterprise customers.

The launch does not mean Google has stopped backing Gemini. It means a company already on Google Cloud no longer has to leave that environment, or manage a separate bill, because one workflow works better on an OpenAI model. It can now test GPT models beside Gemini and Anthropic's Claude in the same platform.

That matters because enterprise AI is no longer a single-model decision. A support team may need a fast model for sorting routine requests and a stronger one for complex complaints. Google Cloud's pitch is 




that Model Garden lets teams match each job to the right model.

Credit: Google Cloud Vertex AI Model Garden interface, via IT Pro Magazine

What was actually announced

Google Cloud's announcement, published on October 21, 2026, names three OpenAI models rolling out in private preview: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. Customers in the preview can build with the models through Google Cloud's APIs and connect local Codex CLI and Desktop workflows directly to Google's Gemini platform for inference.

Model Garden is described as a unified platform with more than 200 first-party, third-party, and open models. With OpenAI added, Google Cloud says it is now the only hyperscale cloud provider offering Gemini, Claude, and GPT on one platform — its own claim about breadth of choice, not an independent ranking of quality.

Access is limited: this is a private preview for select enterprise customers, not a general launch, with an interest form for others. Reporting on OpenAI's announcement says general availability is planned in the coming weeks, but no specific date was published in Google Cloud's announcement, so that should be treated as a reported plan, not a guaranteed date.

The three models, and how Google Cloud describes them

Google Cloud gives each model a different job. GPT-6 Astra is positioned for demanding work across coding, computer use, professional tasks, and scientific research — the kind of work where a weaker model might produce a plausible but unreliable answer.

GPT-6.1 Sol is described as offering near-Astra intelligence at lower cost across agentic coding, computer use, and professional work. Google Cloud is not claiming Sol is identical to Astra. It is presenting Sol as a middle option for teams that want strong agentic and professional performance without paying for the top tier on every request.

GPT-6 Luna is described as a fast, cost-efficient model for everyday work at scale: drafting, classifying, summarizing, routing, and other routine tasks where speed and cost matter more than maximum reasoning power.

These are Google Cloud's descriptions, not independent benchmarks. Preview teams should test all three on their own documents, code, and queries before choosing.

Why the billing change may matter more than the model list

For many enterprises, the most important detail is not which models were named. It is how they are bought. Google Cloud says OpenAI models are transacted directly through Google Cloud Marketplace. All usage appears on a single Google Cloud invoice, draws down existing cloud commitments, and counts toward Flexible Savings Plans across model providers.

That changes procurement: finance teams see AI usage alongside other cloud spending, existing commitments can cover OpenAI usage, and teams can switch models on performance, latency, and cost without, in Google Cloud's words, architectural lock-in or separate vendor contracts.

Google Cloud's announcement does not publish model-specific preview prices, and reporting notes pricing was not disclosed. Teams should confirm the preview price, metering, and pricing at general availability before scaling a pilot into production.






Credit: Google Cloud Vertex AI Model Garden interface, via Klu.ai 


Data, security, and what is still coming soon

Security is where careful buyers should slow down and read the fine print. Google Cloud sets out three commitments: strict data perimeters, isolated access boundaries, and unified auditing and identity.

Google Cloud says customer prompts, context, and outputs remain entirely within the customer's designated data perimeter and are never used to train foundation models — a commitment Help Net Security also reports. Businesses still need to establish who can access company information, how activity is recorded, and what happens to data after a request is processed.

Two related protections are listed as coming soon, not available at launch: Zero Data Retention options and Confidential Compute. Google Cloud says OpenAI models run on dedicated infrastructure, queries pass through hardened, isolated gateways, and every interaction operates under an authenticated identity, with VPC Service Controls protection and unified audit logging.

That "available now" versus "coming soon" distinction is the key point for security teams. If a workload requires Zero Data Retention or Confidential Compute today, the preview may not yet meet that requirement.

Codex, Gemini for business, and Antigravity

Alongside the models themselves, Google Cloud announced that preview customers can build through its APIs and connect local Codex CLI and Desktop workflows to its Gemini platform for inference, with Google Cloud processing the model requests.

Two further integrations are coming soon: OpenAI models in Gemini for business and in Google Antigravity, Google's agent-driven coding platform. Until they arrive, teams should not assume OpenAI models are available inside every Google Cloud AI product — the confirmed route is Model Garden, through APIs and Codex workflows.

What this means for the cloud AI race

TechTarget puts the deal in wider context: OpenAI ended its cloud exclusivity with Microsoft and debuted its models on AWS in April 2026, and this move makes OpenAI models available on all major cloud providers. Reporting on OpenAI's announcement also states that Microsoft remains OpenAI's primary cloud partner — an important correction to any claim that Google Cloud has replaced Microsoft.

For Google Cloud, the benefit is retention: as TechTarget analyst Sid Nag of Tekonyx put it, the partnership strengthens Google's ability to retain AI spending regardless of which model customers select. For OpenAI, the benefit is reach inside clouds customers already use.

For customers, the benefit is choice — though, as Futurum Group analyst David Nicholson told TechTarget, more options can leave enterprises unsure which model to use, even if most would rather have that choice than not. The practical answer is testing: a short pilot with the same tasks, evaluation criteria, and cost tracking will tell a team more than any vendor description.

What businesses should do next

If your company already uses Google Cloud, a preview is most useful when you have a specific workload to test — an internal coding assistant, a document pipeline, or a support-triage system — and a clear way to judge whether Astra, Sol, or Luna improves results or lowers cost.

Before production use, confirm in writing: the preview price and its change at general availability, the regions where the models will run, whether Zero Data Retention or Confidential Compute is required and available, and how audit logs and access controls work for OpenAI workloads — plus who owns support when something goes wrong.






Credit: Google Cloud Vertex AI Model Garden interface, via DevelopersIO 


Key Takeaways

  • Google Cloud announced on October 21, 2026, that OpenAI models are rolling out in private preview to select enterprise customers in Model Garden.
  • The three models are GPT-6 Astra for demanding coding, computer-use, professional, and research work; GPT-6.1 Sol for near-Astra intelligence at lower cost; and GPT-6 Luna for fast, cost-efficient everyday work at scale.
  • Usage is bought through Google Cloud Marketplace, appears on a single Google Cloud invoice, draws down existing cloud commitments, and counts toward Flexible Savings Plans.
  • Google Cloud says prompts, context, and outputs stay within the customer's data perimeter and are never used to train foundation models — but Zero Data Retention and Confidential Compute are listed as coming soon, not available at launch.
  • Preview access covers Model Garden APIs and Codex CLI and Desktop workflows; OpenAI models in Gemini for business and Google Antigravity are coming soon.
  • No model-specific preview prices or a specific general-availability date were published in Google Cloud's announcement, so both should be confirmed before scaling beyond a pilot.

Google Cloud opening its platform to OpenAI models is not the end of the Google–OpenAI rivalry — it is a sign that the fight has moved to who provides the most useful, governable, and affordable place to run many models. For enterprises, that is good news, provided they test carefully, read the security fine print, and choose by measured results rather than headline alone.

Sources

Comments

Post a Comment