OpenAI Models and Codex Are Coming to Google Cloud: What It Costs and How to Use Your Existing Credits

OpenAI Models and Codex Are Coming to Google Cloud: What It Costs and How to Use Your Existing Credits

OpenAI will offer its frontier models and its Codex coding agent through Google Cloud "in the coming weeks," according to Reuters — the latest move in OpenAI's push to sell its technology to large enterprises on whatever cloud they already use. The models will be available through Google Cloud's Model Garden and its Cloud Marketplace, and enterprises will be able to pay for them using the Google Cloud spending commitments and credits they already have. For regulated industries, the models will run under Google Cloud's own access controls and data-governance tooling — the setup companies need before they can put AI agents to work on sensitive data.

OpenAI's frontier models and the Codex coding agent are set to appear inside Google Cloud's Model Garden and Marketplace in the coming weeks. (Illustration: CloudEmythos)

What Happened

The short version: OpenAI is bringing its best models and its Codex coding agent to Google Cloud, ending any remaining idea that OpenAI belongs to one cloud. Reuters reported on October 21, 2026 that the models will roll out through Google Cloud in the coming weeks, sold and delivered the way Google sells other enterprise AI — through Model Garden, its model catalog, and through Google Cloud Marketplace, its procurement channel.

The context matters. OpenAI's models were once distributed exclusively through Microsoft, its largest backer. That exclusivity ended in April, and Amazon Web Services was added in between. Google Cloud is now the latest hyperscaler to carry them, so all three biggest cloud platforms can sell you the same frontier AI — and competition shifts from "who has the models" to price, billing convenience, and governance.

One honesty note: the core news comes from Reuters' reporting. The model names and private-preview rollout below come from Google Cloud's own announcement, as summarized by Unite.AI — read them as the vendor's framing until the Marketplace listings are live.

What Exactly Is Coming: Model Garden and Marketplace, Explained Simply

Two Google Cloud storefronts are involved: Model Garden is where your developers find and call the models; the Marketplace is where your company buys them. If you do not live inside Google Cloud, here is what each one does.

Model Garden is Google Cloud's catalog of AI models — a single shelf where enterprises browse, test, and deploy models from Google and from third parties, with one set of APIs and one security model. Google Cloud says its catalog already holds more than 200 models. Adding OpenAI's models there means a developer working in Google Cloud can call an OpenAI model the same way they call any other model in the garden, instead of maintaining a separate integration with OpenAI or another cloud.

Google Cloud Marketplace is the procurement layer: it is how an enterprise buys third-party software and AI with its normal Google Cloud purchasing process — one vendor relationship, one invoice, one approval chain. For large companies, this is not a minor detail. Buying a new AI vendor directly can mean months of legal, security, and finance review. Buying the same technology through a marketplace the company already uses can be a purchase order.

Codex is OpenAI's coding agent — the tool that writes, edits, and reviews code for developers. Its arrival on Google Cloud means enterprise development teams can get the agent through the same procurement and governance channel as the models, rather than contracting for it separately.

How Buying With Your Existing Google Cloud Credits Works

If your company has a Google Cloud spending commitment, you will be able to pay for OpenAI models and Codex out of that commitment instead of setting up a new billing relationship with OpenAI. This is the part of the deal finance teams will care about most.

Large enterprises typically sign committed-use agreements with a cloud provider: they promise to spend a set amount over one to three years in exchange for discounts. Every dollar of that commitment has to be spent somewhere, and unspent commitments are wasted money. When a third-party product is sold through Google Cloud Marketplace in a way that draws down the commitment, the enterprise can effectively pay for OpenAI's models with budget it has already allocated — and the purchase counts toward meeting the commitment rather than adding a new line item on top of it.

Practically, that removes three frictions at once: no new vendor onboarding, no separate OpenAI invoice, and no new procurement review if your organization already approves Marketplace purchases. For a committed Google Cloud customer, OpenAI's models just became one of the easiest frontier models to buy.

On exact pricing: per-model rates will follow the Google Cloud Marketplace listings when they go live, and neither Reuters' report nor Google Cloud's announcement summary pins down final prices. Anyone quoting you a specific per-token price for OpenAI models on Google Cloud today is guessing. Budget from the Marketplace listing when it appears, and check whether your commitment terms cover Marketplace purchases at the rate you expect.

Codex, OpenAI's coding agent, is part of the Google Cloud rollout — aimed at enterprise development teams that buy through their cloud. (Illustration: CloudEmythos)

Codex on Google Cloud vs. Buying Direct From OpenAI

The agent is the same; what changes is the wrapper — billing, identity, and governance. Buying Codex through Google Cloud routes it through your existing cloud contract, your organization's access controls, and Google Cloud's data-governance tooling. Buying direct from OpenAI means a separate commercial relationship, separate admin, and OpenAI's own enterprise controls.

For regulated industries — finance, healthcare, insurance, government suppliers — the Google Cloud route has a specific advantage Reuters highlighted: the models are offered under Google Cloud's access controls and data-governance tooling for agentic workflows. In plain terms, the company can apply the same rules it already uses to decide who can touch what data, keep audit trails in the systems it already runs, and let agents work inside those boundaries. For a bank or a hospital, that governance layer is often the difference between a pilot and a refusal.

Buying direct still has its case: it can mean earlier access to new OpenAI features, a simpler relationship with the model's maker, and one less intermediary when something breaks. The sensible enterprise question is not "which is better?" but "which wrapper does my compliance and finance team already trust?" If that answer is Google Cloud, this launch removes the main reason to wait.

Which Model Should You Pick? Astra vs. Sol vs. Luna

Google Cloud's announcement names three OpenAI models, in a simple ladder: Astra for maximum capability, Sol for near-flagship work at lower cost, and Luna for fast, high-volume tasks. These names and descriptions come from Google Cloud's announcement, as summarized by Unite.AI, and the rollout is described as a private preview for select enterprise customers first.

• GPT-Astra — the flagship. Google Cloud positions Astra as the top model: the one you point at your hardest reasoning, your most valuable agent workflows, and the tasks where a wrong answer costs more than the model does. If you are evaluating one model for frontier performance, this is it.
• GPT-6.1 Sol — near-flagship, cheaper. Sol is framed as the workhorse tier: most of the flagship's ability at a lower price. For most production workloads — document processing, support agents, code assistance, analysis — this is usually where the cost-to-quality balance lands, and it is the tier most enterprises will end up standardizing on.
• GPT-6 Luna — fast and cheap. Luna is the volume tier: quick responses and low cost for high-throughput jobs like classification, summarization, extraction, and first-pass drafting, where you call the model thousands of times and speed matters more than peak reasoning.

A practical way to choose: prototype on Astra to learn what "good" looks like on your task, then test Sol and Luna against that bar. Many teams find the cheaper tier is good enough for most of their traffic, and they keep the flagship only for the hard cases. Google Cloud also claims, in its announcement, that it is now the only hyperscaler offering all three leading closed-source model families alongside the 200+ models in Model Garden — a claim about breadth that, if it holds, makes Google Cloud the one catalog where enterprises can compare the top models side by side under one contract.

What This Means for the Microsoft–OpenAI Relationship

It means the exclusive era is fully over, and Microsoft now competes to sell OpenAI's models like everyone else. Microsoft remains OpenAI's biggest backer and Azure remains a major home for OpenAI workloads, but the distribution monopoly that once defined the partnership ended in April. AWS followed, and Google Cloud — Microsoft's fiercest cloud rival — now sells OpenAI's frontier models and coding agent directly to enterprises.

For OpenAI, the logic is straightforward: enterprises are divided among the three clouds, and selling through only one leaves most of the market to rivals. For Microsoft, the pressure moves to what is still distinctively theirs — Azure integration depth, the Copilot products built on top, and the partnership's commercial terms. For buyers, the effect is healthy: when the same models are on every shelf, the clouds compete on price, credits, and tooling, and the customer keeps the leverage.

FAQ

When will OpenAI models and Codex be available on Google Cloud?
"In the coming weeks," according to Reuters' October 21, 2026 report. Google Cloud's announcement describes a rollout beginning in private preview for select enterprise customers, so general availability through Model Garden and the Marketplace will follow the preview phase.

Can I pay for OpenAI models with my existing Google Cloud credits?
Yes. Enterprises can pay using their existing Google Cloud spending commitments and credits, because the models are sold through Google Cloud Marketplace. Instead of a separate OpenAI contract and invoice, the purchase draws on the Google Cloud budget your company has already committed.

Which OpenAI models are coming to Google Cloud?
Google Cloud's announcement names three: GPT-Astra, the flagship; GPT-6.1 Sol, a near-flagship at lower cost; and GPT-6 Luna, a fast, cheaper model for high-volume work — alongside the Codex coding agent. The names and preview status come from Google Cloud's announcement, as summarized by Unite.AI.

How much will OpenAI models cost on Google Cloud?
No final prices have been published. Pricing will follow the Google Cloud Marketplace listings when they go live. Expect the three named models to sit at different price points matching their tiers — flagship highest, Luna lowest — but treat any specific per-token figure quoted before the listings appear as speculation.

Why does selling through Google Cloud matter for regulated companies?
Because the models come under Google Cloud's access controls and data-governance tooling for agentic workflows. A bank, hospital, or insurer can apply its existing rules for who accesses what data, keep its audit trail in one place, and still use OpenAI's models and Codex — which is often the condition for getting an AI deployment approved at all.

Sources

Sources: Reuters, "OpenAI to sell its top models on Google Cloud in latest enterprise push" (October 21, 2026) — reuters.com; Google Cloud announcement details (model names, private-preview rollout, Model Garden catalog claims) as summarized by Unite.AI. Model names, tier descriptions, and preview status are attributed to Google Cloud's announcement; pricing had not been published at the time of writing.