Key takeaways

  • OpenAI announced the partnership on October 21, 2026. Google Cloud says GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out now in private preview to selected enterprise customers; OpenAI says general availability is planned in the coming weeks.
  • Selected customers can build through Google Cloud APIs and connect local Codex CLI and Desktop workflows to the Gemini platform for inference, but neither announcement publishes endpoint URLs, request schemas, region coverage, quotas or service levels.
  • Purchases will transact through Google Cloud Marketplace, appear on one Google Cloud invoice and draw down eligible GCP commitments; Google says usage also counts toward Flexible Savings Plans across model providers.
  • No numeric Google Cloud rates are published. OpenAI’s direct-API Standard rates range from $0.10/$0.50 per million input/output tokens for Luna to $10/$50 for Astra, but those numbers must not be treated as Google Cloud preview prices.
  • Treat data-perimeter, training-use, dedicated-infrastructure, identity and audit statements as Google claims requiring account and contract verification. Zero Data Retention and Confidential Compute are explicitly coming soon, not preview features to assume today.
01

Three GPT-6 models enter a selected-customer preview

OpenAI announced on October 21, 2026 that its frontier models and Codex workflows are coming to Google Cloud’s Model Garden. Google Cloud names the first three models as GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna and says they are rolling out now in private preview to selected enterprise customers. OpenAI separately says general availability is planned in the coming weeks. This is a real access expansion, but it is not yet an unrestricted generally available service.

The two first-party pages have a date inconsistency worth retaining. OpenAI displays October 21. Google Cloud’s BlogPosting metadata also says datePublished 2026-10-21, while the rendered Google Cloud article displayed October 22 during AccessAllGPT’s October 21 review. The product facts align; the visible Google dateline should not be used to manufacture a later availability date.

02

The preview covers Astra, Sol and Luna

Google Cloud positions GPT-6 Astra for demanding coding, computer use, professional tasks and scientific research; GPT-6.1 Sol as a lower-cost near-Astra option; and GPT-6 Luna for fast, high-volume everyday work. OpenAI’s live references identify the direct API models as gpt-6-astra, gpt-6.1-sol and gpt-6-luna. Each accepts text and image input, returns text, and documents a 1,050,000-token context window with up to 128,000 output tokens.

Those OpenAI identifiers and limits establish the underlying public model contracts, not the complete Google Cloud serving contract. The announcements do not publish Google Cloud endpoint names, payload examples, supported regions, per-project limits, preview quotas, tool parity, snapshot policy or model lifecycle terms. Confirm each field inside the selected Google Cloud project before porting a workload.

03

Codex can route inference through the Gemini platform

Google Cloud says selected customers can start building with the models through its APIs and connect local Codex CLI and Desktop workflows directly to the Gemini platform for inference. OpenAI describes Codex workflows as part of the partnership. Neither page provides the setup command, authentication flow, policy boundary or feature matrix, so “Codex on Google Cloud” should not be read as proof that every Codex surface, tool or hosted workflow is included.

OpenAI models are also promised for Google Antigravity and the Gemini Agent, but both announcements describe those integrations as coming soon. Keep preview API access, local Codex routing and future product integrations as three separate availability states in procurement and architecture records.

04

Marketplace procurement is clearer than usage pricing

OpenAI says customers will purchase the models through Google Cloud Marketplace and can apply eligible GCP commitments. Google Cloud adds that usage appears on one Google Cloud invoice and counts toward Flexible Savings Plans across model providers. That can simplify procurement and commitment utilization, but it does not reveal the unit economics.

Neither launch page publishes Google Cloud per-token prices, tool-call charges, preview credits, minimum commitments, support costs or regional premiums. OpenAI’s current direct Standard rates are $10 input and $50 output per million tokens for Astra, $2 and $10 for Sol, and $0.10 and $0.50 for Luna; long-context and processing-mode modifiers also apply. Use those only as direct-API reference points. Require a Google Cloud price sheet and a measured bill before comparing hosts.

05

Some governance controls are claims and others are future work

Google Cloud says prompts, context and outputs remain inside the customer’s designated data perimeter and are not used to train foundation models. It also describes dedicated Google Cloud infrastructure, hardened isolated gateways, authenticated identities, VPC Service Controls and unified audit logging. These are vendor statements about the intended service boundary, not an independent audit by AccessAllGPT.

Two prominently named controls are not available to assume today: Google says Zero Data Retention options and Confidential Compute are coming soon. Preview evaluators should record actual retention, logging, support access, key management, regional routing, subprocessors and deletion behavior for their account and contract. A roadmap sentence does not clear a mandatory data or compliance gate.

06

What cloud AI buyers should do next

Request access only for a named workload and compare the same frozen cases through the existing provider and the Google Cloud preview. Capture the exact model and host identifiers, region, request format, supported tools, latency distributions, accepted outcomes, quota failures, logs, retained data, invoice line items and commitment treatment. Test Codex routing separately from direct API calls because the operational and permission boundaries may differ.

Adopt only after access, price, contract and controls are evidenced for the account; constrain the preview to reversible non-critical workloads; wait when ZDR, Confidential Compute, regions, quotas or lifecycle terms are mandatory and unresolved; reject a migration justified only by commitment drawdown or vendor claims without workload evidence.

07

Copy-ready Google Cloud OpenAI preview record

Use this for one model, one Google Cloud project and one workload before approving a preview or migration.

Entries stay in this browser tab and are not submitted to AccessAllGPT. Blank responses are copied as [Unresolved].

Interest submitted; selected or not selected; preview or GA; project and organization; approval date; account owner; expiry and support contact.

Google Cloud endpoint and API version; model ID and snapshot; region; modalities; context and output limits; tools; quotas; rate limits; SLA and lifecycle terms.

CLI or Desktop version; authentication; project and identity; local data sent; tools available; policy controls; logs; fallback; feature gaps.

Google Cloud unit rates; tool charges; regional or mode premiums; credits; GCP commitment and FSP treatment; support; measured invoice; cost per accepted outcome.

Data perimeter; training and improvement use; retention; ZDR availability; logging; support access; subprocessors; deletion; encryption; Confidential Compute status.

Frozen cases; current-host baseline; accepted outcomes; critical failures; latency distributions; token and tool use; quota behavior; reviewer effort.

IAM; VPC Service Controls; audit logs; budget alerts; incident route; regional failure; model change; rollback and exit drill.

Adopt, constrain, wait or reject; evidence links; unresolved gates; rollout ceiling; owner; review date; GA, price, model, contract or control changes that force retesting.

Attach dated console evidence, contracts, price sheets, request and response samples, evaluation results, audit logs, invoices and rollback evidence. Mark vendor statements separately from observed account behavior.

Primary sources

Browse the publication-wide evidence index →

  1. OpenAI models are coming to Google CloudOpenAI · Reviewed: October 21 announcement date; Model Garden and Codex partnership scope; Marketplace transaction path; GCP commitment drawdown; Google Antigravity and Gemini Agent roadmap; general-availability timing; Microsoft cloud-partner statement · Retrieved · Supports: OpenAI says its frontier models and Codex workflows are coming to Google Cloud Model Garden, purchases will transact through Google Cloud Marketplace and can draw down GCP commitments, general availability is planned in the coming weeks, and Microsoft remains OpenAI’s primary cloud partner.
  2. Google Cloud expands model choice and flexibility, bringing OpenAI models to customersGoogle Cloud · Reviewed: Publication metadata and rendered dateline; three named models; private-preview availability; API and Codex connection statements; Marketplace billing and Flexible Savings Plans; data-perimeter, training-use, infrastructure, identity and audit claims; coming-soon ZDR and Confidential Compute · Retrieved · Supports: Google Cloud names GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, says they are rolling out in private preview to select enterprise customers, and says selected customers can use APIs and connect local Codex CLI and Desktop workflows. Google also describes Marketplace billing, governance controls and two security features that remain forthcoming.
  3. GPT-6 Astra model referenceOpenAI Developers · Reviewed: Direct OpenAI API model ID; model positioning; input and output modalities; context and output limits; reasoning levels; Standard token prices; long-context multiplier; endpoints, features and tools · Retrieved · Supports: OpenAI documents gpt-6-astra for text and image input and text output with a 1,050,000-token context window, 128,000-token maximum output, and direct-API Standard prices of $10 input and $50 output per million tokens. These are not published Google Cloud preview rates.
  4. GPT-6.1 Sol model referenceOpenAI Developers · Reviewed: Direct OpenAI API model ID; positioning; modalities; context and output limits; reasoning levels; Standard token prices; long-context and processing-mode modifiers; Responses and Chat Completions support · Retrieved · Supports: OpenAI documents gpt-6.1-sol for text and image input and text output with a 1,050,000-token context window, 128,000-token maximum output, and direct-API Standard prices of $2 input and $10 output per million tokens. These are not published Google Cloud preview rates.
  5. GPT-6 Luna model referenceOpenAI Developers · Reviewed: Direct OpenAI API model ID; positioning; modalities; context and output limits; reasoning levels; Standard token prices; long-context and processing-mode modifiers; Responses and Chat Completions support · Retrieved · Supports: OpenAI documents gpt-6-luna for text and image input and text output with a 1,050,000-token context window, 128,000-token maximum output, and direct-API Standard prices of $0.10 input and $0.50 output per million tokens. These are not published Google Cloud preview rates.

Limitations

This is a public-source launch briefing, not hands-on testing, a security assessment, a contract review or a price comparison. AccessAllGPT did not receive Google Cloud private-preview access; submit the access form; obtain Google Cloud or OpenAI credentials; call GPT-6 Astra, GPT-6.1 Sol or GPT-6 Luna through Google Cloud; connect Codex CLI or Desktop; inspect endpoints, payloads, regions, quotas, latency, tools, logs, invoices, commitment drawdown, Flexible Savings Plan treatment, retention, training use, data perimeters, gateway isolation, VPC Service Controls, Zero Data Retention or Confidential Compute; or reproduce any capability claim. Direct OpenAI model specifications and prices are not proof of Google Cloud preview terms. The first-party sources control the claims and can change.

Disclosures

AccessAllGPT received no OpenAI or Google Cloud account, preview place, model access, API credits, GCP commitments, Flexible Savings Plan, contract, price sheet, private briefing, review, payment or compensation for this article. OpenAI and Google did not sponsor, review or endorse it. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with OpenAI, Google, Google Cloud or Microsoft. Publication-wide relationships are listed on the disclosures page.

Further AccessAllGPT guidance

  1. OpenAI Rolls Out GPT-6 and Intelligent UI Across ChatGPT
  2. GPT-6.1 Sol: Price, Context and Migration Gates
  3. A Model Guide for the GPT-6 Family
  4. Google Cloud Introduces Gemini, a Universal Work Agent
  5. AI API Data Retention and Residency: Set the Procurement Gates
  6. AccessAllGPT Research methodology
  7. Evidence standards
  8. Publication disclosures

Continue the research

Get evidence-led updates for teams making production AI decisions.