• Google Cloud's own announcement of October 21, 2026 confirms GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out in private preview to select enterprise customers.
• Billing runs through Google Cloud Marketplace, so usage lands on one invoice and draws down existing commitments. No per-token prices appear in the announcement.
• Zero data retention options and Confidential Compute are described as coming soon, so they cannot be relied on in a compliance assessment today.
• Google says it is the only hyperscale provider offering all three leading closed-source model families. The Decoder widened that to "only major provider", which is a different claim.
• Neither the announcement nor the report includes benchmarks, so any quality or cost comparison against your current model has to come from your own evaluation.

Google Cloud has confirmed that OpenAI's GPT-6 models are available to some customers, and the confirmation is more limited than the headlines suggest. In an announcement dated October 21, 2026, Google says three models are in private preview for select enterprise customers. It publishes billing and data-handling language but no prices, regions or benchmarks, and it labels two of its protections as coming soon.

What Google actually announced

Google's announcement states that GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out now in private preview to select enterprise customers in Model Garden, with more coming soon. It describes Astra as the model for demanding coding, computer-use and research work, Sol as near-Astra intelligence at lower cost, and Luna as the fast, cost-efficient option. Interested customers are directed to a sign-up form, so access is by request, not by default.

Billing and contracts are the real change

The most consequential detail is commercial, not technical. Google says OpenAI models are transacted directly through Google Cloud Marketplace, so usage appears on a single Google Cloud invoice, draws down existing cloud commitments and counts toward Flexible Savings Plans. For procurement teams, that may remove a separate vendor contract and security review. The announcement publishes no per-token prices, so the cost advantage, if any, cannot be assessed yet.

Data-handling promises: some live, some pending

Google says prompts, context and outputs remain within the customer's designated data perimeter and are never used to train foundation models. It also says interactions run under authenticated identity with VPC Service Controls and audit logging. Zero data retention options and Confidential Compute are both described as coming soon. A governance review should therefore record only the live statements as commitments and treat the pending ones as unproven until they appear in contract terms.

The exclusivity claim needs careful reading

Google's wording, in a quote credited to Michael Gerstenhaber, is that it is now the only hyperscale cloud provider offering access to all three leading closed-source LLMs. The Decoder's headline says only major provider, which is broader. We did not audit rival catalogues, so this stays a vendor claim. It also describes catalogue breadth, not model quality, and says nothing about whether the models perform better on a given workload.

Verdict: binding soon, not binding now

On the Fine Print scale this is binding soon. The announcement is real and the billing route is concrete, but access is gated, prices are missing, and two protections are future promises. Request preview access if the models fit your roadmap, get retention, region and service-level terms in writing, and run your own evaluation against your current model. No benchmark has been published, so a switch should rest on your own results.