News · 2026-10-22

OpenAI models come to Google Cloud after April’s contract change

OpenAI and Google Cloud announced on October 21 that OpenAI models and Codex workflows are coming to Google Cloud’s Model Garden, with purchasing through Cloud Marketplace. Eligible customers can apply existing cloud spending commitments. General availability is planned in the coming weeks, making this a distribution and procurement change whose full rollout is still ahead.

Key facts

• Google names three models: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna.
• OpenAI and Google Cloud announced the arrangement on October 21, 2026.
• An April 27 amendment had already allowed OpenAI to serve products through other clouds.
• The primary source is OpenAI’s Google Cloud announcement.

For an enterprise, choosing a model involves more than comparing answers. The organization may already have a cloud contract, approved purchasing route, audit system and network boundary. Adding another supplier can require repeating that work. Buying a model through an established marketplace can reduce those administrative obstacles even when the model itself is familiar. That buyer-friction interpretation is an inference from the announced purchasing path, rather than a measured customer outcome.

Google Cloud’s announcement names the models and describes dedicated infrastructure, authenticated identities, network controls and unified audit logging. Google says prompts, context and outputs remain within the customer’s designated data perimeter and are not used to train foundation models. These are the provider’s stated controls; the announcement is not an independent audit or a substitute for the customer’s contractual review.

A useful analogy is a company adding another manufacturer’s equipment to an already approved purchasing catalog. The equipment can compete on its own merits, while the catalog makes it easier to order, pay for and govern. Model Garden becomes the selection surface, Marketplace handles the transaction, and existing spending commitments can cover eligible purchases. That combination changes the commercial route to the model without necessarily changing how intelligent it is.

The timing matters. OpenAI says general availability is expected “in the coming weeks.” The dossier’s checked launch text does not establish a private preview restricted to selected enterprise customers, so that detail should not be added. Nor should the offering be renamed Vertex AI from memory: the announcements identify Model Garden and the Gemini platform. Accurate product names prevent readers from looking for access in the wrong place or assuming a familiar deployment contract applies unchanged.

The Microsoft story also began earlier than this launch. OpenAI’s April 27 partnership summary says it may serve all its products through any cloud. It makes Microsoft’s license to OpenAI intellectual property non-exclusive through 2032 while retaining Microsoft as the primary cloud partner. New products ship first on Azure when Microsoft can support them, with the stated exception if Microsoft cannot and chooses not to support the necessary capabilities.

That differs from the October 2025 summary, which described Azure application-interface exclusivity and exclusive intellectual-property rights under its stated conditions. The Google distribution agreement uses permission established in April. Reporting it as the event that broke exclusivity would merge a contract amendment with a later commercial implementation and erase a meaningful part of the timeline.

The complete underlying contract is not public in those summaries. Readers can verify the announced rights and conditions, but should avoid inferring every operational or financial detail. In particular, availability through another cloud does not establish equivalent prices, identical service terms, identical release timing or automatic portability for every Codex workflow. Each would require its own published specification or contractual evidence.

Google describes itself as the only hyperscale provider offering the three leading closed-model families. That is competitive positioning rather than an independently established ranking. The Hacker News discussion debated marketplace strategy and commitment accounting. Its comments help explain what practitioners are considering, but do not supply audited evidence of savings or prove which company benefits most financially.

There are explicit remaining roadmap items. Google describes zero-retention and confidential-computing capabilities as coming soon; readers should not treat them as already active guarantees. Procurement teams should distinguish announced controls from delivered ones and compare terms for their actual workload. The same discipline applies to inference economics: the quoted token bill is one part of a larger operational cost. This launch matters because model choice is becoming easier to fit into existing enterprise arrangements, while availability and governance details still determine the usable product.

Primary source, verified: read the paper →

Ground Truth. (2026, October 22). OpenAI models come to Google Cloud after April’s contract change. Ground Truth. https://groundtruth.day/news/openai-google-cloud-model-garden-distribution.html

@misc{groundtruth:openai-google-cloud-model-garden-distribution,
  title  = {OpenAI models come to Google Cloud after April’s contract change},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/openai-google-cloud-model-garden-distribution.html}
}

Topics: openai · google-cloud · enterprise · distribution · cloud