OpenAI Models Land on Google Cloud: GPT-6 Astra, Sol, and Luna in Private Preview

Google Cloud is now selling OpenAI models. On October 21, 2026, Google Cloud announced that three OpenAI models, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, are entering its Model Garden in a private preview for select enterprise customers. The company says it is now the only hyperscale cloud provider offering all three leading closed-source large language models, Google’s Gemini, Anthropic’s Claude, and OpenAI’s GPT, on one unified enterprise AI platform, alongside more than 200 first-party, third-party, and open models.

The move completes a long shift away from Microsoft’s exclusive hold on OpenAI’s cloud distribution. Under a renegotiated agreement announced in April 2026, OpenAI gained the freedom to sell its models on any cloud provider, and GPT-6 models are now available on Azure, AWS, and Google Cloud. Microsoft remains OpenAI’s primary cloud partner, but “first” no longer means “only.”

Three models for three workloads

The announcement frames the launch around matching each workload to the right model rather than using one model for everything. Google Cloud describes the three models as follows:

GPT-6 Astra: designed for demanding work across coding, computer use, professional tasks, and scientific research.
GPT-6.1 Sol: near-Astra intelligence at lower cost across agentic coding, computer use, and professional work.
GPT-6 Luna: a fast, cost-efficient model for everyday work at scale.

Google Cloud has not published its own Model Garden rates for these models. For reference, TechTimes cites OpenAI’s direct API prices at $10 per million input tokens and $50 per million output tokens for Astra, $2 and $10 for Sol, and $0.10 and $0.50 for Luna. That is a 100-fold difference between Astra’s output price and Luna’s, which is why Google Cloud pushes customers to match workloads to tiers: reserve Astra for tasks that need maximum reasoning depth, and use Luna for high-volume, less precision-critical work. Whether Google Cloud’s Marketplace rates match, exceed, or undercut the direct OpenAI API remains unknown.

One invoice, and where the data stays

The commercial mechanics may matter as much as the model lineup. Usage is transacted through Google Cloud Marketplace, so all OpenAI model spend appears on a single Google Cloud invoice, draws down existing cloud commitments, and counts toward Flexible Savings Plans (FSP) across model providers. Starting now, customers in the preview can call the models through Google Cloud’s APIs and connect local Codex CLI and Desktop workflows to the Gemini platform for inference. OpenAI’s coding agent, Codex, is also listed as coming to Google Cloud as part of the broader rollout, and OpenAI models are slated to arrive in Gemini for business and Google Antigravity. General availability is expected within weeks, with sign-ups already open.

On the security side, Google Cloud says customer prompts, context, and outputs remain entirely within the customer’s designated data perimeter and are never used to train foundation models. The models run on dedicated Google Cloud infrastructure, and queries are routed through hardened, isolated gateways. Two stronger protections, Zero Data Retention (ZDR) options and Confidential Compute, are listed as coming soon rather than available in the initial preview. Enterprise security teams can additionally use VPC Service Controls to keep inference traffic inside their Google Cloud project boundary, so requests never traverse the public internet.

Why it matters

For enterprises already committed to Google Cloud, the pitch is straightforward: three model families, one invoice, one security and compliance baseline, and no separate vendor contracts with OpenAI, Anthropic, and Google individually. Google Cloud’s ability to swap models by cost, latency, and performance without architectural lock-in is the real selling point for teams running agentic workloads.

There is a tradeoff worth noting. Consolidating all three model families onto one provider runs counter to the multi-cloud risk management that many enterprise architecture guidelines recommend. If Google Cloud’s infrastructure degrades or its terms change, access to all three AI families is affected at once. As TechTimes puts it, the billing consolidation argument is strong if you already have substantial Google Cloud committed spend, but thin if you maintain a deliberately multi-cloud footprint.

The industrial logic for Google is simple: cloud compute is fungible. Every token of GPT-6 inference served on Google hardware is compute consumed and billed through Google Cloud, regardless of whose weights run. Box is one of the first named customers. “By using OpenAI models through Google Cloud, we will be able to give customers access to OpenAI’s frontier AI models alongside Google Cloud’s security and compliance capabilities,” said Yashodha Bhavnani, VP of AI Products at Box.

The larger question is competitive. Google Cloud now hosts and bills for its biggest AI rival’s most capable models, betting that cloud market share and ecosystem depth outweigh the cost of strengthening OpenAI’s distribution. Whether that calculus holds as Gemini competes more directly with Astra-tier capabilities will be one of the more interesting dynamics to watch over the next few product cycles.