# OpenAI Models on Google Cloud: What Developers Must Know > Source: > Published: 2026-10-21 17:09:20+00:00 OpenAI’s frontier models landed on Google Cloud today — and for enterprises sitting on committed GCP spend, the billing mechanics matter more than the model names. As of October 21, [GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are rolling out in private preview on Google Cloud’s Model Garden](https://openai.com/index/openai-models-google-cloud/), with Codex agentic workflows bundled in. Most developers cannot access them yet, but when general availability lands, Google Cloud becomes the only hyperscale platform running OpenAI, Anthropic’s Claude, and Google’s own Gemini under one roof. ## What Is Actually Available Three models are in the rollout. GPT-6 Astra is the top-tier option, aimed at demanding coding, computer use, and research workloads. GPT-6.1 Sol sits one tier below at a lower price point — expect this to be the workhorse model for most enterprise teams. GPT-6 Luna handles high-volume, latency-sensitive tasks where cost efficiency matters more than raw capability. Codex agentic workflows are included, and there is a detail worth flagging: starting today, developers can also connect Codex CLI and Desktop directly to Google’s Gemini platform for inference. That means the Codex interface, running Gemini models underneath. Cross-vendor agent tooling is quietly becoming real. The catch: this is private preview only. Google and OpenAI have not published a GA date beyond “coming weeks.” If you are planning a production migration, nothing has changed today except the direction of travel. ## The Billing Angle Nobody Is Talking About Here is what enterprise engineering leaders should actually care about: OpenAI models on Google Cloud Marketplace draw down existing GCP committed spend. All usage appears on a single Google Cloud invoice and counts toward Flexible Savings Plans across model providers. What this means in practice: a team sitting on a 0M annual GCP commitment — common at mid-market and enterprise scale — can now route that committed budget to OpenAI models without opening a new procurement cycle or negotiating a separate contract with OpenAI. For organizations where new vendor approvals take months, this is the actual unlock. The models are not new. The purchasing path is. ## Google Cloud Now Has All Three LLM Families The competitive picture shifted today. [Google Cloud now hosts Gemini natively, Claude via Vertex AI (generally available), and OpenAI models via Model Garden](https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-brings-openai-models-to-customers) — making it the only hyperscale provider to offer all three leading closed-source LLM families on one platform. | Platform | OpenAI | Claude / Anthropic | Google Gemini | |---|---|---|---| | Azure | Yes (GA, first) | Partial (AI Studio) | No | | AWS Bedrock | Yes (GA, Jun 2026) | Yes (GA) | No | | Google Cloud | Private preview | Yes (GA) | Yes (native) | Google’s positioning here is deliberate. Rather than betting everything on Gemini being the best model, Google Cloud is betting on model neutrality — that enterprises want to pick the right model for each task and consolidate billing, governance, and infrastructure in one place. The winners of the cloud AI era may not be the labs with the best models; they may be the platforms that make switching between models cheapest. ## Why This Took Until October Context matters. [Microsoft and OpenAI formally ended their cloud exclusivity in April 2026](https://interestingengineering.com/ai-robotics/microsoft-openai-partnership-restructuring), restructuring a deal that had made Azure the only hyperscaler legally permitted to host OpenAI’s proprietary models. Azure kept first-to-market rights on new models, but the exclusivity wall came down. [GPT-5.5 and Codex became generally available on Amazon Bedrock in June 2026](https://www.aboutamazon.com/news/aws/bedrock-openai-models), with the GPT-5.6 family following in July. Google is the last major cloud to complete the lineup — but the only one that can now claim all three tier-1 LLM families. ## What to Do Now If you are running on GCP and have existing OpenAI API spend, request private preview access through Google Cloud Marketplace. Compare actual pricing once published — AWS Bedrock confirmed parity with OpenAI’s first-party rates; Google has not yet stated its pricing structure. If you are not yet on GCP, this announcement does not change your calculus today. Wait for GA, then compare governance features and pricing against AWS Bedrock, where OpenAI has been stable since June. The broader shift is clear: the era of one-cloud, one-model-provider lock-in is over. What developers should be optimizing for now is model-agnostic infrastructure — the ability to swap models without changing your deployment pipeline. Google Cloud just made that easier for GCP customers. Whether the pricing justifies a workload migration depends on numbers that are not public yet.