Google Cloud gets OpenAI’s GPT-6 models in private preview

OpenAI and Google Cloud announced a partnership on October 21 that puts OpenAI’s frontier models and Codex agentic workflows into Model Garden, Google’s catalogue of hosted models. Three models, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, are rolling out now in private preview to select enterprise customers.

The changes that matter in practice are how the models get billed and where inference runs. OpenAI’s announcement covers the partnership itself. Google’s own launch post has the details: the model lineup, the billing terms, the security controls, and the fact that Codex already works against Google’s infrastructure.

Google describes the three models as tiers. GPT-6 Astra is the top one, “designed for demanding work across coding, computer use, professional tasks, and scientific research.” GPT-6.1 Sol offers “near-Astra intelligence at lower cost” for agentic coding, computer use and professional work. GPT-6 Luna is the fast, cost-efficient option for everyday work at scale. Google says customers can swap between these and the rest of Model Garden based on performance, latency and cost “without architectural lock-in or separate vendor contracts.”

Codex is the piece developers can use today. Google says customers can now build with the models through its APIs and connect their local Codex CLI and Desktop workflows directly to the Gemini platform for inference. The Codex client stays on your machine, and the model calls go to OpenAI models running on Google Cloud instead of OpenAI’s own endpoints.

Billing is where finance teams will feel it. The models are sold through Google Cloud Marketplace, so all usage lands on a single Google Cloud invoice. Companies with Google Cloud spend commitments can draw them down with OpenAI usage, and Google says that usage also counts toward Flexible Savings Plans across model providers. A company that has pre-committed a budget to Google can spend part of it on GPT-6 without signing a separate contract with OpenAI.

Both companies pitch the deal to regulated industries. OpenAI’s post says deploying AI there “requires control over who can access sensitive data and how it is used.” Google lists the controls:

• Customer prompts, context and outputs stay inside the customer’s designated data perimeter and are never used to train foundation models.
• OpenAI models run on dedicated Google Cloud infrastructure, with queries routed through hardened, isolated gateways so data in transit stays inside Google Cloud. Google says third-party model providers have no unauthorized access to the customer’s environment.
• Every agent and model interaction runs under an authenticated identity, with VPC Service Controls protection and unified audit logging.

Two items on that list are still promises. Zero Data Retention options, under which no copy of prompts or outputs is kept after a request, are listed as “coming soon,” and so is Confidential Compute on the dedicated infrastructure. The regulated-industry pitch arrives before two of its own parts.

The two companies also describe availability differently. OpenAI says general availability is planned “in the coming weeks” and asks customers to request access. Google describes a private preview for select enterprise customers, with more models “coming soon,” and points to an interest form. Both say OpenAI models will later reach Google Antigravity, Google’s agentic coding environment, and Google’s business agent. OpenAI calls that product the Gemini Agent, while Google’s post says Gemini for business. Neither gives a date. OpenAI frames the Google channel as giving enterprises “a path from evaluation to broader adoption.”

Google is using the launch to claim it is now the only hyperscale cloud provider offering access to all three leading closed-source LLMs, next to more than 200 first-party, third-party and open models. Its agent products were already multi-vendor: the Gemini agent routes each job between Gemini models and Anthropic’s Claude. Adding OpenAI fills in the third name.

None of this was possible under the original Microsoft deal. It gave Microsoft exclusive cloud rights to OpenAI’s models, so AWS, Google Cloud and Oracle Cloud could not license or serve them. The amended agreement Microsoft announced on April 27 lets OpenAI serve all its products to customers on any cloud provider. Microsoft keeps a license to OpenAI IP for models and products through 2032, and that license is now non-exclusive.

OpenAI’s post ends by noting that Microsoft remains its primary cloud partner. Under the April terms, that status comes with a condition: OpenAI products ship first on Azure, unless Microsoft cannot and chooses not to support the necessary capabilities.