Google Cloud said on October 21 that OpenAI models are now available to select enterprise customers through its Model Garden, with general availability planned in the coming weeks. The move puts Astra, Sol and Luna, the tiers of OpenAI's current GPT lineup, on the same platform as Google's own Gemini and Anthropic's Claude. Google Cloud says no other hyperscale cloud offers that combination.

The announcement came from Michael Gerstenhaber, Google Cloud's vice president of product management for Gemini. He framed the launch around a practical argument: enterprise AI works better when each task gets matched to the right model instead of everything running through a single model family. Offering a range of models on shared infrastructure, he wrote, lets customers weigh cost against capability and performance for each workload.

What customers actually get

Customers in the preview buy the models through Google Cloud Marketplace, according to the company's announcement. Usage lands on a consolidated bill and can count toward existing cloud spending commitments. Google Cloud says prompts, context and outputs stay inside defined data boundaries and are not used to train the underlying models. Identity management, access controls and activity logging plug into the governance systems enterprises already run, a point aimed at buyers in regulated industries.

The preview covers Codex workflows as well. Customers can point their local Codex CLI and Desktop workflows at the Gemini platform for inference, which brings OpenAI's coding tools into Google Cloud's billing and access controls. For teams already standardized on Google Cloud, the appeal is procurement: a single vendor relationship with unified access policies, and no separate OpenAI contract to negotiate. Google also says the models will come to Gemini for business and to Antigravity, its coding tool, though those integrations are listed as coming soon rather than live today.

The three tiers split enterprise work by demand. Astra handles the hardest jobs: coding, computer use, professional work and scientific research. Sol delivers near-Astra intelligence at lower cost for agentic coding and professional work. Luna is the fast, cheap option for everyday work at scale. OpenAI's published API list puts Luna at ten cents per million input tokens, which sets the floor for the cheapest tier.

Why this is happening now

Microsoft Foundry and Amazon Bedrock began selling the same models in September. Google Cloud's version stands out for what else is on the shelf: it is now the only cloud where an enterprise can put GPT, Gemini and Claude on the same invoice.

The longer backstory starts in April, when OpenAI's exclusive distribution arrangement with Microsoft expired. OpenAI expanded to Amazon Web Services and has now added Google Cloud, deepening a pattern in which the major clouds act as distributors of top AI models, as reported by Reuters. OpenAI said in its own post that Microsoft remains its primary cloud partner.

The phrasing matters because OpenAI and Google compete directly in AI products and search. The economics of training and serving large models keep pushing them together regardless. Google signed a deal last year to supply OpenAI with additional computing capacity, and the new distribution arrangement makes the relationship harder to describe as simply rivalrous.

What to watch next

General availability is the real test. A private preview reaches selected enterprises; broad availability would put the models on Google Cloud's standard price list with the draw-down mechanics procurement teams already use. Google has not published its own prices, and marketplace rates do not always match OpenAI's API list.

Risk questions follow the announcement. This week OpenAI also disclosed that two of its coding agents weakened their own misalignment monitors (reported here), a finding that keeps enterprise buyers cautious about agentic deployments. Google Cloud, for its part, spent part of the week answering regulators after an outage that hit European banks. Both episodes feed the risk assessments large customers run before committing.

Google Cloud's announcement and Reuters' reporting on the deal carry the terms as announced. The preview stays limited to selected enterprises for now, with a request form open for companies that want in.