OpenAI Models Arrive on Google Cloud: What Multi-Cloud AI Means for Swiss SMEs

On October 21, 2026, OpenAI and Google Cloud announced that OpenAI's models are coming to Google's Model Garden, starting with a private preview for selected enterprise customers. It is the biggest structural AI story of the week, and not because of a new benchmark. For the first time, the leading OpenAI models are on offer from all three major US clouds. For a Swiss SME that already pays one of them, that changes how you buy, secure and switch AI.

What Was Announced

Based on the announcements and secondary coverage:

• Models: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are reported as the initial lineup, with the Codex coding agent following.
• Status: private preview for selected enterprise customers via an interest form. OpenAI says general availability is planned within weeks. Google has not given a date.
• Billing: usage through the Google Cloud Marketplace is reported to count toward an existing Google Cloud spend commitment, giving one contract and one invoice.
• Positioning: OpenAI's statement says Microsoft remains its primary cloud partner. Microsoft's licence became non-exclusive in late April 2026, which opened the door for this deal.

Treat third-party details, especially per-token prices quoted by blogs, as unconfirmed until they appear on Google's own pricing page.

Why This Matters More Than Another Model Launch

Until this spring, using OpenAI models through a hyperscaler essentially meant Azure. Now the same model family is reachable through the cloud you may already use for storage, identity and analytics. Three practical effects follow.

1. Less Procurement Friction

If your IT contract is with Google Cloud, adding a GPT model no longer needs a new vendor onboarding, a new security review and a separate invoice. For a ten-person IT team, that can be the difference between "pilot in a week" and "stuck in procurement for a quarter."

2. A Real Exit Option

Vendor concentration was the lesson of our article on the Pentagon dropping Anthropic Claude. When one supplier suddenly becomes unavailable, restricted or more expensive, the ability to move to another model within the same cloud account is worth far more than a small price difference. Claude, Gemini and now GPT models side by side in one platform is a hedge, provided your code can actually switch.

3. Competition on Price and Reliability

Cloud marketplaces make comparison easier. We already argued in our cost-per-task comparison of Sonnet 5.5 and GPT-6.1 Sol that sticker prices say little. With every model one API away, the useful number is cost per accepted result, measured on your own data.

What Does Not Change

• Model behaviour. A model on a different cloud is still the same model, with the same strengths and failure modes. Hosting does not fix reliability.
• Your governance duties. Where prompts and documents are processed, who can see logs and how long data is retained remain your responsibility. Our piece on the governance gap of shadow AI agents still applies.
• Lock-in at the application layer. Using three models from one cloud is not the same as being portable. If your prompts, tool definitions and evaluation sets are written for one provider's quirks, you are still locked in.

A Practical Checklist for Swiss SMEs

1. Check your region. Confirm in which region the preview models run and whether a Swiss or EU region is offered. Do not assume it matches your current setup.
2. Review data terms. Read the data-processing and retention terms for third-party models in Model Garden. They can differ from Google's own models. Check fit with the Swiss FADP and, for EU customers, the GDPR.
3. Abstract the model call. Keep the model name and provider in configuration, behind one internal interface, so a swap is a one-line change.
4. Keep an evaluation set. 50 to 200 anonymised real examples per workflow let you compare candidates in an afternoon.
5. Wait for general availability for production. Private preview is for testing. Do not move live workflows onto it.
6. Do not migrate for its own sake. If your current setup works and costs are acceptable, the right response may be to do nothing now and re-evaluate when pricing is published.

The Bottom Line

OpenAI on Google Cloud is not a reason to rebuild your AI stack. It is a reason to make sure you can. The market is moving toward a world where every major model is available on every major cloud, and the companies that benefit are those who can compare and switch without a project. Build that ability now, while nothing is forcing you to.

Not sure how portable your AI setup is? Talk to our team. We can review one workflow and show you what switching would take.

Sources: OpenAI models come to Google Cloud (daily.dev) · OpenAI Models Come to Google Cloud: What Is in Preview (Digital Applied) · OpenAI ends Microsoft exclusivity, brings GPT to Google Cloud (NewsBytes) · OpenAI Ends Microsoft Exclusivity (NeuralWired)