OpenAI Comes to Google Cloud: GPT-6 Models and Codex Land in Model Garden

Two of the fiercest rivals in AI are now doing business together. On Wednesday, October 21, 2026, OpenAI announced that its frontier models and Codex agentic workflows are coming to Google Cloud's Model Garden — the catalog where Google Cloud customers pick foundation models — with general availability planned in the coming weeks. Customers can request access now.

This is a distribution deal, not a technology merger. But the mechanics of it matter more than the headline: enterprises will be able to buy OpenAI models through Google Cloud Marketplace, pay on a single Google Cloud invoice, draw down existing cloud spend commitments, and have usage count toward Google's Flexible Savings Plans. If you are a CFO, that sentence was the actual announcement.

What was actually announced

According to Google Cloud's official blog post, the partnership brings OpenAI's newest models to Model Garden, the company's catalog of first-party, third-party and open foundation models. The initial lineup is three GPT-6 models, all rolling out now in private preview to select enterprise customers.

The deal has four moving parts:

• Model access: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in Model Garden, with API access for private-preview customers.
• Marketplace billing: OpenAI models transacted directly through Google Cloud Marketplace — all usage on one Google Cloud invoice.
• Codex workflows: OpenAI's agentic coding tools available alongside the models, with local Codex CLI and Desktop workflows able to connect directly to the Gemini platform for inference during the preview.
• Google's own tools: OpenAI models "will soon be available" in Google Antigravity and the Gemini Agent, giving users model-choice flexibility inside Google's development environments.

"This partnership is an important step in making our most capable models and Codex available where customers are already building," said OpenAI president Greg Brockman. Google Cloud CEO Thomas Kurian said the company is "focused on giving customers a broad selection of models, including those from OpenAI."

The three GPT-6 models arriving

Google Cloud describes the initial models as a full-spectrum enterprise lineup. Here is how the company positions each one — these are Google's descriptions, not independent test results:

The "right model for every workload" framing is Google's pitch, and it is worth reading skeptically: vendors always say their three models cover everything. But the structure itself is the industry standard now — a flagship, a value tier and a speed tier — and OpenAI's lineup slots neatly into it.

The billing part is the real story

Most coverage of this deal leads with the models. I would lead with the invoice. Here is the fine print that actually changes enterprise behavior:

• One bill. Because OpenAI models are transacted through Google Cloud Marketplace, usage appears on the single Google Cloud invoice a customer already pays.
• Commitment drawdown. Enterprises with Google Cloud spend commitments can apply them to OpenAI model purchases. For large shops that pre-bought cloud capacity, this converts a sunk commitment into AI budget.
• Flexible Savings Plans. OpenAI usage counts toward Google's Flexible Savings Plans across model providers, the same mechanism that discounts other cloud consumption.
• No separate vendor contract. Google Cloud's blog explicitly says teams can swap models "without architectural lock-in or separate vendor contracts."

And the crucial honesty check: no per-token pricing, no revenue-sharing terms and no exact GA date have been disclosed. Axios, via secondary reporting, noted the announcement included none of these figures. Reuters, in its October 21 report, added that cloud companies typically take a percentage of sales made through their platforms — meaning Google almost certainly takes a cut of every OpenAI token sold through Marketplace, though neither company confirmed terms. Until prices are published, anyone budgeting for this should treat it as a procurement channel, not a priced product.

Codex comes along too — and that matters

Codex is OpenAI's agentic coding stack: the tool that plans and executes software development tasks rather than just autocompleting code. Its inclusion is not decoration. The enterprise AI spend that is actually growing right now is agentic workflows — software teams buying compute to let agents do multi-step work — and OpenAI's blog post, per TeslaNewsWire's summary, frames the whole partnership around "complex agentic workflows" and modernizing the software development lifecycle.

During the private preview, customers can connect local Codex CLI and Desktop workflows directly to the Gemini platform for inference. The longer-term play is Codex inside Google Antigravity and the Gemini Agent — putting OpenAI's coding agent inside Google's own development tools, which is a striking sentence to type about two companies competing in AI products and search.

Why this matters: the Microsoft-exclusive era is over

Context is what turns this from a press release into a trend. Reuters reports that OpenAI's exclusive distribution arrangement with Microsoft expired in April 2026. Since then, OpenAI has expanded distribution to Amazon Web Services and now Google Cloud. CNBC notes the move puts OpenAI's models on all three of the top U.S. cloud providers.

The industry shift here is bigger than OpenAI. Hyperscalers are becoming the primary commercial channels for frontier AI labs — PYMNTS and Reuters both note the pattern. Google, for its part, claims to be "the only hyperscale cloud provider offering access to all three leading closed source LLMs," alongside more than 200 other models. That is Google's own claim, not an independently verified ranking, and it is doing exactly the job it is supposed to do: positioning Google Cloud as the neutral platform while Google and OpenAI compete everywhere else.

It is also worth noting the two companies were already entangled: Google struck a cloud deal last year to give OpenAI more computing capacity. Compute economics push rivals into cooperation even while they compete in products — a dynamic we covered in our analysis of DeepSeek open-sourcing its DeepInfer inference engine, where the industry is simultaneously renting distribution and giving away infrastructure.

What this does not change — yet

A few things to keep straight before anyone rewrites their architecture diagrams:

• Microsoft stays first. OpenAI's own post states that Microsoft remains its primary cloud partner. This is expansion, not defection.
• Preview means limited. Only select enterprise customers get the private preview. Everyone else waits for general availability, which is "in the coming weeks" — a phrase that can mean anything from days to months.
• Security claims are vendor claims. Google Cloud says customer prompts, context and outputs remain within defined data boundaries and are not used to train the underlying models, with identity management and audit logging integrated. That is the right architecture for regulated industries, but it is Google's description of its own product; compliance teams will verify it themselves.
• Antigravity and Gemini Agent integration is "soon," not now. The models are coming to Google's dev tools, but no date was given.

FAQ

When will OpenAI models be generally available on Google Cloud?

Neither company gave a date. The announcement says general availability is planned "in the coming weeks," with a private preview rolling out to select enterprise customers now. Customers can request access through Google Cloud in the meantime.

How much do OpenAI models cost on Google Cloud?

No per-token pricing has been disclosed. What is known: purchases are transacted through Google Cloud Marketplace, appear on your Google Cloud invoice, can draw down existing spend commitments, and count toward Flexible Savings Plans. Expect pricing at or near OpenAI's direct rates — the value proposition is billing convenience, not necessarily lower prices.

What is Google Cloud Model Garden?

It is Google Cloud's catalog of foundation models — first-party Google models, third-party models like OpenAI's, and open models — that customers can deploy on Google's infrastructure with unified billing and governance. Google says the catalog now holds more than 200 models.

Does this mean OpenAI is leaving Microsoft?

No. OpenAI explicitly says Microsoft remains its primary cloud partner. The Google deal is distribution expansion following the expiry of Microsoft's exclusive arrangement in April 2026 — OpenAI is now on all three major U.S. clouds.

Is Codex included in the Google Cloud deal?

Yes. Codex agentic workflows are part of the announcement, and private-preview customers can connect local Codex CLI and Desktop workflows to the Gemini platform for inference. OpenAI models are also slated to arrive in Google Antigravity and the Gemini Agent.

Key takeaways

• OpenAI's frontier models are coming to Google Cloud's Model Garden, announced October 21, 2026 — private preview now, general availability in the coming weeks.
• The lineup is GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna — a flagship, a value tier and a speed tier — plus Codex agentic coding workflows.
• The billing mechanics are the real product: Marketplace transactions, a single invoice, drawdown of existing cloud commitments, and Flexible Savings Plan credit.
• No per-token pricing has been disclosed — treat this as a procurement channel with announced mechanics, not a fully priced product, until Google publishes rates.
• The Microsoft-exclusive era is definitively over: OpenAI's models now reach all three top U.S. clouds, and hyperscalers are becoming the main distribution channel for frontier AI.

For years, the AI race looked like a duel — OpenAI versus Google, models versus search, ChatGPT versus Gemini. On October 21, 2026, the duel added a line item: Google Cloud now sells OpenAI's models. The rivalry is not over. It just learned to invoice.