Google Cloud has become the only major cloud offering all three leading closed frontier model families — its own Gemini, Anthropic's Claude, and now OpenAI's GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna — in one catalogue, on one bill. Here is what launched, what it costs (and what Google has not yet told us about cost), and which teams should care.
Key takeaways
- Three OpenAI models, private preview: GPT-6 Astra (flagship for coding, computer use, professional tasks and research), GPT-6.1 Sol ("near-Astra intelligence at lower cost"), and GPT-6 Luna (fast, cost-efficient, for everyday work at scale) are rolling out now in Google Cloud's Model Garden to select enterprise customers, with an interest form open.
- One bill, existing commitments: Usage is transacted through Google Cloud Marketplace, lands on a single Google Cloud invoice, draws down existing cloud commitments and counts toward Flexible Savings Plans (FSP) across model providers.
- Data terms are the headline feature: Google states prompts, context and outputs stay inside your designated data perimeter and are never used to train foundation models. Zero Data Retention options and Confidential Compute are listed as "coming soon", not available today.
- No public pricing yet: Neither Google's announcement nor independent coverage publishes per-token prices. Google claims "market-leading price-performance" — a vendor claim, unverified. Treat any budget modelling as provisional until rates are published.
- Coming soon, not today: OpenAI models in Gemini for Business and in Google Antigravity (its agent-powered coding tool) are announced as "coming soon"; Google has not specified which of the three models Antigravity will get.
What actually launched
Starting October 22, enterprise customers in the private preview can build with the three models through Google Cloud APIs. Developers can also connect local Codex CLI and Codex Desktop workflows to Google's platform for inference — the models run on Google's infrastructure, not OpenAI's. They join a Model Garden catalogue of more than 200 first-party, third-party and open models.
| Model | Positioned for | What we know |
|---|---|---|
| GPT-6 Astra | Demanding work: coding, computer use, professional tasks, scientific research | Flagship tier; expected to be the most expensive of the three to run (per Ars Technica) |
| GPT-6.1 Sol | Agentic coding, computer use, professional work | "Near-Astra intelligence at lower cost" (Google's positioning) |
| GPT-6 Luna | Everyday work at scale | Fast and cost-efficient tier |
A note on names, because it will confuse procurement teams: Google's official naming makes only Sol a "6.1" model — Astra and Luna are GPT-6. Some coverage groups all three under "GPT 6.1"; use Google's names in contracts and SKU discussions.
Why this matters: the multi-model shelf is now a billing feature
The strategic point is not that GPT models exist on another cloud — it is that Google is selling substitution. Customers can select and swap models dynamically based on performance, latency and cost, without separate vendor contracts. For an enterprise already committed to Google Cloud spend, GPT usage now burns down the same commitment and the same Flexible Savings Plan as Gemini usage. That removes the two frictions that usually keep teams single-vendor: procurement and committed-spend accounting.
Context worth keeping straight: OpenAI notes Microsoft remains its primary cloud partner, and Google is simultaneously shipping its own Gemini 4 Argon — so this is a distribution deal layered on top of an ongoing rivalry, not an alliance. Google's compute now strengthens the two rivals of its own Gemini team; whether that trade stays comfortable depends on how much enterprise demand the unified shelf captures.
The money part (TCO)
Pricing verified as of Q4 2026: not yet published. No per-token rates for Astra, Sol or Luna on Google Cloud appear in Google's announcement or in independent reporting as of October 22, 2026. What is confirmed:
- Billing route: Google Cloud Marketplace, single invoice.
- Commitment drawdown: usage counts against existing Google Cloud commitments and toward Flexible Savings Plans across model providers — so the effective discount depends on your existing FSP tier, not on a model-specific price list.
- Cost ranking (qualitative only): Astra most expensive, Sol positioned below it, Luna cheapest tier — Google's positioning, not independently benchmarked pricing.
Practical advice: do not migrate workloads on price until Google publishes rates and you can compare cost per completed task (including retries and effort settings) against your current provider. The confirmed savings today are administrative — one contract, one invoice, one commitment pool — not per-token.
Data and security terms
Google lists: strict data perimeters (prompts, context and outputs remain in your designated perimeter, never used to train foundation models); isolated access boundaries on dedicated Google Cloud infrastructure with hardened gateways; unified identity, auditing and VPC Service Controls across model interactions. Two items are explicitly future, not current: Zero Data Retention options and Confidential Compute are both "coming soon". If ZDR is a hard requirement for your compliance review, this preview does not satisfy it yet.
Verdict — who should act, who should wait
- Enterprise teams already committed to Google Cloud: Act now. Join the preview, route a bounded workload (internal coding assistance or document processing) through Model Garden, and measure cost per completed task across Astra/Sol/Luna versus your Gemini baseline. The billing integration alone justifies the experiment.
- Enterprises on Azure or AWS: Wait for published pricing. The offer is a switching-cost argument, not yet a price argument; without rates there is nothing concrete to renegotiate with.
- Independent developers and small teams: Wait. This is an enterprise private preview — no self-serve access or pricing has been announced. Your existing API routes remain the practical option today.
FAQ
Which OpenAI models are available on Google Cloud?
GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, rolling out in private preview to select enterprise customers in Model Garden from October 22, 2026.
How much do the GPT-6 models cost on Google Cloud?
Google has not published per-token pricing yet. Usage is billed through Google Cloud Marketplace and counts toward existing commitments and Flexible Savings Plans.
Is my data used to train models?
Google states customer prompts, context and outputs remain within your designated data perimeter and are never used to train foundation models. Zero Data Retention options are announced as coming soon.
Will GPT models come to Google Antigravity?
Yes — Google says OpenAI models are coming soon to Antigravity and Gemini for Business, but has not specified which models or a date.
Does this change OpenAI's relationship with Microsoft?
No. OpenAI notes Microsoft continues to be its primary cloud partner; the Google deal is an additional distribution channel.
Sources (checked October 22, 2026): Google Cloud Blog, "Google Cloud expands model choice and flexibility, bringing OpenAI models to customers" (Oct 22, 2026); Ars Technica, "OpenAI's latest AI models are coming to Google Cloud and Antigravity" (Oct 22, 2026); The Decoder, "Google Cloud now offers OpenAI models alongside Gemini and Claude" (Oct 21, 2026).
Comments (0)