Google Cloud launched OpenAI models in its model garden on 21 October 2026. GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna arrived in private preview for selected enterprise customers. Google claims it is now the only hyperscaler offering all three leading closed language models.
Key takeaways
- Three OpenAI models: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna
- The model garden holds over 200 first-party, third-party and open models
- Billing runs through Google Cloud Marketplace, on a single invoice
- Microsoft remains OpenAI's primary cloud partner
Three models, three cost tiers
The split of roles is explicit. Astra targets the hardest work: coding, computer use, professional tasks and scientific research. Sol offers near-Astra intelligence at lower cost, and Luna is the fast, cheap model for everyday work at scale.
| Model | What it is for |
|---|---|
| GPT-6 Astra | the hardest work: coding, computer use, professional tasks, scientific research |
| GPT-6.1 Sol | near-Astra intelligence at lower cost |
| GPT-6 Luna | the fast, cheap model for work at scale |
The platform allows swapping models mid-flight, with the choice driven by performance, latency and cost, and no architectural lock-in or separate vendor contracts. Models run entirely on Google infrastructure, and local Codex CLI and Codex Desktop workflows can connect to the Gemini platform for inference.
Billing and data perimeters
Transactions go through Google Cloud Marketplace, so usage lands on one invoice, draws down existing cloud commitments and counts toward Flexible Savings Plans.
On security, Google states that prompts, context and outputs stay inside the designated data perimeter and are never used to train foundation models. OpenAI models run on dedicated infrastructure, with queries routed through isolated gateways. Zero Data Retention and Confidential Compute are announced but not live yet.
The diagram follows Google Cloud's stated claims about query flow and billing. It does not show the gateway's internal architecture or how the data perimeter is enforced.
What it changes in the cloud lineup
The model garden?model garden: a catalogue of AI models ready to run in the cloud, with no infrastructure to maintain already offered Gemini, open models — Mistral, Llama, Gemma — and Claude. Adding GPT completes the set of three closed frontier families at one provider. Antigravity, Google's agentic coding tool, has supported Gemini and Claude so far, with OpenAI models due later. OpenAI notes that Microsoft is still its primary cloud partner.
Why it matters
For an enterprise buyer what counts is not model access but the absence of friction: one invoice, one data perimeter, one audit trail. That shifts competition from the model layer to the billing and compliance layer. For Google it is also an admission that selling cloud beats loyalty to its own Gemini.
What's next
- Access stays in private preview, expanded through an interest form
- Zero Data Retention and Confidential Compute are slated for later
- OpenAI models are due in Gemini for Business and Google Antigravity
