OpenAI Models on Google Cloud: GPT-6 Joins Gemini and Claude

Google Cloud now says it is the only hyperscale cloud offering all three leading closed models, Gemini, Claude and GPT, after adding OpenAI’s GPT-6 family to its model garden. For enterprise teams building agents, the news is less about any one model and more about procurement: OpenAI usage can now land on the same Google Cloud invoice, draw down existing commitments, and sit under the same identity and audit controls as everything else.

What Happened

In Google Cloud’s announcement, Michael Gerstenhaber, VP of product management for Gemini at Google Cloud, lists three OpenAI models arriving in model garden:

They are rolling out in private preview to select enterprise customers, with an interest form for others. Customers can build through Google’s APIs and connect local Codex CLI and Desktop workflows to the Gemini platform for inference. Google says the models will also come to Gemini for business and Google Antigravity.

Why It Matters

The billing detail is the practical change. OpenAI models are transacted through Google Cloud Marketplace, so usage appears on one Google Cloud invoice, counts against existing cloud commitments, and counts toward Flexible Savings Plans. A team with a large Google commit no longer needs a separate OpenAI contract to put GPT-6 into a workflow.

Google also sets out governance terms:

• Data: prompts, context and outputs stay inside the customer’s data perimeter and aren’t used to train foundation models. Zero Data Retention options are listed as coming soon.
• Isolation: OpenAI models run on dedicated Google Cloud infrastructure behind isolated gateways, with Confidential Compute coming soon.
• Identity and audit: every agent and model call runs under an authenticated identity, with VPC Service Controls and unified audit logging.

Box, one early customer quoted by Google, says using OpenAI models through Google Cloud lets it offer them with Google Cloud’s security and compliance capabilities. The “coming soon” items matter for regulated teams, who should not assume ZDR or confidential compute are available in the preview.

What This Means for Agent Builders

Multi-model agent design gets easier when every model shares one bill and one set of access controls. Google’s pitch is swapping models by workload, cost and latency without architectural lock-in. The Decoder notes the tension underneath: Google is spending compute to serve the biggest rival of its own Gemini team.

Stack Archive covered the earlier break in exclusivity in OpenAI on AWS, and the price gaps between these models in GPT-6.1 Sol versus Astra.

Source: Google Cloud Blog

Content created with AI assistance and reviewed for accuracy.