Google Cloud Puts OpenAI’s GPT-6 Models Beside Gemini

Google Cloud began a private preview of OpenAI models on October 21, 2026, listing GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna in its Model Garden. It said it is now the only hyperscale cloud with all three leading closed models, billed on one Google invoice beside Gemini.

Google Cloud Adds OpenAI Models to Model Garden

The listing is a distribution deal, not a merger of the two labs. OpenAI still sells ChatGPT and the API under its own name, and Google still trains Gemini. What changed is the storefront: GPT-6 inference can now run on dedicated Google Cloud machines and show up on a Google Cloud bill.

Michael Gerstenhaber, vice president of product management for Gemini at Google Cloud, wrote that customers can swap models by performance, latency, and cost without a separate vendor contract. That is the pitch to shops that already keep data, identity, and spend on Google and did not want a second AI paperwork pile.

Greg Brockman, OpenAI’s president and co-founder, framed the same move as putting models where the builders already sit.

Google Cloud is a leading platform for developers and businesses building AI agents, writing software, and putting AI to work. This partnership is an important step in making our most capable models and Codex available where customers are already building, giving them even more choice in how they use frontier AI as we grow the agentic ecosystem together.

Greg Brockman, President and co-founder, OpenAI

Choice here is a polite word for substitution. A team can keep Gemini for one job and route the next job to Astra or Sol without leaving Google’s gateways, logs, or procurement file.

The Only Hyperscaler With All Three Closed Models

Google Cloud’s post says the company now offers all three leading closed models, alongside more than 200 first-party, third-party, and open models on one enterprise platform. The closed trio in that sentence is Gemini, Claude, and GPT. Google is advertising that its own flagship shares a catalog with the two labs built to beat it.

Today, we're bringing OpenAI models to Google Cloud's model garden.

We're now the only hyperscale cloud provider offering access to all three leading closed source LLMs—alongside 200+ other models—on a unified enterprise platform → https://t.co/OWExjjWBB7 pic.twitter.com/VwCeIIepJ3

— Google Cloud (@googlecloud) October 21, 2026

Eight days earlier, on October 13, 2026, the same Google Cloud blog launched Gemini 4 Argon for the hardest business jobs, from long finance work to agentic coding in Google Antigravity. Google set Argon’s $2 and $10 rate per million input and output tokens as an introductory price through November 30, 2026, then $4 and $20 after that.

Those intro figures match OpenAI’s public list for Sol. Google did not publish Model Garden rates for GPT-6, so no one should treat the OpenAI API card as the Google bill. The comparison still sits in the open: Argon arrived as Google’s answer to hard work, and eight days later Google put Sol on the same shelf at the same public price point.

We are focused on giving customers a broad selection of models, including those from OpenAI, so they can choose the right model for any given job, helping to deliver optimal quality and lower costs.

Thomas Kurian, Chief Executive Officer, Google Cloud, in OpenAI’s announcement

Kurian’s line is the quiet concession. If Gemini always won on quality and cost, Google would not need to sell the rival to keep the customer on Google Cloud.

GPT-6 Astra, Sol, and Luna in the Preview

Google named three SKUs for the first wave, each aimed at a different spend band. Astra is the heavy model. Sol is the near-Astra option Google describes as cheaper for agent work. Luna is the fast daily driver.

THE THREE GPT-6 MODELS ON GOOGLE CLOUD

OpenAI’s public Sol and Luna API prices are the figures in that table. Sol and Luna shipped in September 2026 at 50% below their GPT-5.6 promotional rates, which is OpenAI’s comparison, not a Google Cloud discount. Marketplace charges on Google Cloud have not been posted, and they may differ once general availability arrives in the coming weeks.

The useful split for a buyer is still the one Google wrote down. Astra is the model you reach for when the job is ugly. Sol is the one you reach for when the job is ugly and the token bill matters. Luna is the one you point at high-volume chores that should not wake Astra.

What Unused Google Cloud Spend Can Buy Now?

Companies with leftover Google Cloud commitments can draw those dollars down against OpenAI usage, because the models sell through Google Cloud Marketplace and land on a single Google Cloud invoice. That turns idle committed spend into GPT-6 budget without opening a new vendor file, which is the part of the deal that moves money, not just model names.

HOW THE BILL RUNS

• One invoice: OpenAI usage posts on the existing Google Cloud bill instead of a separate OpenAI contract.
• Commit drawdown: Enterprises can apply current Google Cloud spend commitments to those purchases.
• Savings plans: The usage also counts toward Flexible Savings Plans across model providers, per Google Cloud.
• No extra contract: Google says teams can swap models without architectural lock-in or a second vendor agreement.

That billing path is why the listing has teeth inside Google shops and less bite everywhere else. A company that already prepaid Google can try Astra on the leftover commit. A company that prepaid Azure still has to justify a new cloud relationship.

The live question is whether those Google shops actually move inference. A catalog line can sit unused for a year while production traffic stays on Gemini, or on OpenAI’s own API, because swapping a model in a demo is cheaper than swapping it in a regulated workflow. The procurement door is open. The production door is a separate fight.

Codex Today, Antigravity and the Gemini Agent Next

The first workflow that is actually live is Codex, not the chat window. From October 21, customers in the preview can call the models through Google Cloud APIs and point local Codex CLI and Desktop sessions at the Gemini platform for inference. That is a practical bypass for teams that already live in Codex and want Google’s network, identity, and spend controls around the run.

Google says OpenAI models will later show up in Gemini for business, Google Antigravity, and the Gemini Agent. Antigravity already runs Gemini and Claude. Adding GPT would give a coding agent three closed labs in one desktop, which is the product version of the catalog claim, and it is not in the preview.

WHAT IS LIVE AND WHAT IS STILL COMING

1. October 13, 2026: Gemini 4 Argon becomes available on Google Cloud, in Gemini for business, and in Google Antigravity.
2. October 21, 2026: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna roll out in private preview on Model Garden, and Codex CLI and Desktop can connect to the Gemini platform for inference.
3. Coming weeks: OpenAI plans general availability of the Google Cloud listing.
4. Coming soon: OpenAI models in Gemini for business, Google Antigravity, and the Gemini Agent, plus Zero Data Retention options and Confidential Compute.

Until Antigravity ships those models, the Google path is an API and a CLI redirect. That still matters for a shop that wants GPT-6 behind Google’s audit logs. It does not yet make GPT a first-class twin of Argon inside Google’s own agent app.

Prompts Stay Inside the Customer’s Perimeter

Google Cloud is selling this as a way for regulated teams to use OpenAI without handing the prompts to a loose third-party pipe. Prompts, context, and outputs stay inside the customer’s designated data perimeter and are not used to train foundation models, Gerstenhaber wrote. Queries route through isolated gateways on dedicated Google Cloud infrastructure, and Google says the model provider gets no unauthorized path into the customer’s cloud.

Identity and logging sit on Google’s side too. Every agent and model call is supposed to run under an authenticated identity, with VPC Service Controls and unified audit logs across the footprint. That is the control story Box bought.

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, enabling them to get more value from their enterprise content.

Yashodha Bhavnani, VP of AI Products at Box, in Google Cloud’s announcement

Two promised controls are still missing. Zero Data Retention is “coming soon,” and so is Confidential Compute. The preview is a perimeter and a dedicated rack, not the full isolation set Google is using as the enterprise close. Buyers who need ZDR as a contract term do not have it on this listing yet.

Microsoft Remains OpenAI’s Primary Cloud Partner

OpenAI’s post still states that Microsoft remains OpenAI’s primary cloud partner. That line is doing a lot of work. It tells Microsoft the training and hosting relationship is not being rewritten in public. It tells Google customers the GPT-6 SKUs are a storefront, not a statement that Azure has been dropped.

The models in this preview still run on dedicated Google Cloud infrastructure and settle on a Google invoice. Primary partner, in that setting, is a hierarchy of compute and contract, not a ban on Google selling the API. OpenAI already needed more than one cloud to serve demand. Google Cloud is now a place to buy the models, not only a place to rent chips.

Gemini 4 Argon still has home-field advantages Google did not give OpenAI: it is already in Gemini for business and Antigravity, with a posted price card. GPT-6 has a private preview, a Marketplace SKU, and a promise to follow into those same tools. Until that promise ships, Google’s “all three” claim is true in the garden and incomplete in the agent window.

The customer who already pays Google can try that garden on money that was going to sit idle. Argon then has to win the next workload with Astra, Sol, and Luna sitting on the same bill.