OpenAI unveils GPT-6 Astra on Google Cloud in private preview

OpenAI and Google Cloud announced this Wednesday (21st) that three OpenAI models will be offered in the Model Garden, the Vertex AI platform's model catalog, in a private preview for selected corporate clients. The models are GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, with general availability expected in "the coming weeks," without a fixed date, according to both companies.

Access currently requires approval: interested parties fill out forms from both companies. Purchases go through the Google Cloud Marketplace, and OpenAI indicated that customers could apply their consumption against existing spending commitments with Google Cloud, according to SiliconANGLE and PYMNTS.

What's included in the catalog

GPT-6 Astra is OpenAI's flagship model, aimed at programming, computer use, professional tasks, and scientific research. GPT-6.1 Sol is described by Google as having "performance close to that of Astra" at a lower cost, while GPT-6 Luna is the quick and economical option for high volume. OpenAI's programming tool, Codex, is also coming to the Google environment, which promises the models for Antigravity, its development environment, "soon," according to 9to5Google.

The announcement leaves gaps that corporate buyers will notice. There is no price per million tokens, no list of regions, no technical specifications, and no general availability date. For those needing data residency in the European Union or Brazil, the absence of regions is a deciding factor on whether the preview serves any purpose beyond testing.

Why channel matters more than model

Thomas Kurian, CEO of Google Cloud, stated that the company wants to provide customers "a wide selection of models, including those from OpenAI," so they can choose the right one for each task, "delivering quality and lower costs." Google, which sells its own Gemini, now also resells the direct competitor within the same platform.

Microsoft remains OpenAI's primary cloud partner, according to OpenAI itself. However, this move extends the company's presence in the three largest cloud services: a comparison by Digital Applied had already listed OpenAI models on AWS and Azure before Google. For OpenAI, the logic is distribution. Large companies commit to multi-year consumption agreements with one cloud provider, and a model that can be paid using that balance fits into the budget without a new contract.

A skeptical reading is also valid. A private preview without published pricing and deadlines is still not a product. Customers expecting to use GPT-6 Astra on Google Cloud in production rely on general availability that neither company has dated. Moreover, the competition between Google and OpenAI in the consumer space, via Gemini and ChatGPT, remains unchanged: the partnership covers corporate sales but does not end competition.

The distinction that matters for the CIO is between model and contract. The choice of model has become a task-specific decision, reviewable every quarter. The choice of cloud still involves a multi-year contract. When cutting-edge models from different labs are available on more than one cloud, the bargaining power returns to the cloud contract, and the model ceases to be a locking argument.

How this reaches each market

In the United States, the immediate effect is on banks, retailers, and media companies that have centralized data in BigQuery and have avoided moving information out of Google Cloud to use OpenAI models. From general availability onward, the call to the model happens within the same billing and governance perimeter.

In Europe, especially in Germany and the United Kingdom, where Google Cloud competes for public sector and industry contracts, the question is the same as for Brazil: in which regions will the models run? A GPT-6 Astra served only from American data centers does not meet data sovereignty requirements that have become standard clauses in these contracts.

In India, integrators operating centers of excellence in Google Cloud, such as TCS, Infosys, Wipro, and Accenture, gain a commercial argument: they can propose architectures with Gemini and GPT on the same platform without forcing the client to switch clouds. In Brazil, where Google Cloud maintains the São Paulo region, the announcement only gains practical weight if that region is included when OpenAI and Google publish the missing details.

The key data to watch in the coming weeks is pricing. If GPT-6 Astra costs the same on Google as on OpenAI's direct API, the advantage is in convenience and budgeting. If it costs less, large clients will begin comparing bills from the three clouds before selecting a model.