OpenAI Google Cloud Partnership: 3 GPT-6 Models Land
On October 21, 2026, OpenAI and Google Cloud announced a partnership that brings three of OpenAI's most advanced AI models into Google Cloud's Model Garden. This is the first time customers of Google's cloud platform can license OpenAI models directly through the platform alongside Google's Gemini models. Both companies confirmed the deal, which was also reported by Ars Technica and CNBC. The three models are GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. OpenAI's Codex tools for carrying out multi-step coding tasks will also be available. This gives businesses another way to buy AI tools through a marketplace that has mainly offered Google's own models and other models with publicly available weights.
Cloud platforms and AI model makers have often worked closely together, making it harder for customers to choose products from different companies. This deal changes that pattern. Google Cloud will not only host a competitor's technology but also sell it, handle billing, and provide its own tools for managing access, data, and security. The announcement comes as businesses ask cloud providers for more AI choices, less dependence on a single supplier, and simpler ways to buy AI products. It also puts more pressure on Microsoft Azure, AWS Bedrock, and Anthropic as they plan their multi-model strategies for 2027 budgets.
What OpenAI and Google Cloud Actually Announced
The main point is simple: OpenAI's advanced AI models and Codex tools for completing multi-step tasks are coming to Google Cloud's Model Garden. Model Garden is Google's catalog of AI models. It lets business customers set up, test, and use Google's own models and models from other companies within their existing Google Cloud environment. According to OpenAI's announcement, customers will be able to use the models directly through Google Cloud to run complex tasks that involve several steps. They can use Google's computing infrastructure and data-management controls without needing a separate contract or another cloud account.
Michael Gerstenhaber of Google Cloud described the deal as a continuation of an existing relationship, not a one-time event. In the Google Cloud Blog, he wrote: “Today, we are expanding that foundation by launching OpenAI models in Google Cloud’s model garden.” OpenAI made a similar point in its announcement, saying that “Google Cloud’s longstanding investment in machine learning and AI has made it a destination for AI workloads and the developers and data scientists building them.”
Google Cloud presents this deal as proof that its platform aims to support many different AI companies, not just Google. The company says it is now the only large cloud provider offering access to all three leading groups of closed-source large language models, along with more than 200 other models in Model Garden. If this claim holds, it would give Google Cloud a wider choice of models than Microsoft Azure or AWS Bedrock, even though those platforms also have their own special partnerships and model offerings.
Private Preview Today, General Availability in the Coming Weeks
The models are not yet available to everyone. GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are first being offered in a private preview to selected business customers. This means Google Cloud and OpenAI are testing the service with a limited group rather than making it available to every Model Garden user at once. Both companies say the models should become generally available in the coming weeks. However, neither company has announced an exact date, and news reports about the deal have not provided one either.
This gradual launch follows a pattern OpenAI has used for its most advanced models on other platforms in 2026. A small group of selected business customers tests the models first and gives feedback about speed, cost, and security controls. The companies can then open access more widely after testing the service with real workloads. For large companies already using Google Cloud, this preview period is a chance to request access, compare the models against their own tests, and plan their budgets before the models become generally available.
Meet the Three Models: Astra, Sol, and Luna
The three OpenAI models coming to Google Cloud are designed for different needs. Each offers a different balance between cost and ability. This suggests that OpenAI expects businesses to use several models for different jobs rather than rely on just one model across an entire company. GPT-6 Astra is the most powerful option and is designed for demanding coding tasks, computer use, and scientific research. Shattered.io previously compared Astra with Claude Opus 5.5 in a Terminal-Bench test. GPT-6.1 Sol is described as offering intelligence close to Astra's at a lower cost. It is aimed at coding agents and professional tasks where price matters as well as ability. GPT-6 Luna is the faster, cheaper choice for everyday tasks that need to be handled in large numbers.
| Model | Positioning | Primary Use Cases | Google Cloud Access Stage |
|---|---|---|---|
| GPT-6 Astra | Top-tier frontier model | Complex coding, computer use, professional and scientific-research tasks | Private preview |
| GPT-6.1 Sol | Near-Astra intelligence at lower cost | Agentic coding, computer use, professional work | Private preview |
| GPT-6 Luna | Fast, lower-cost everyday model | High-volume, everyday work at scale | Private preview |
This range of models also fits with OpenAI's recent pricing changes. Shattered.io separately reported that GPT-6 Sol and Luna launched at about half the price of their predecessors. OpenAI has made similar price cuts across several releases this year as it competes with Google's Gemini models and newer, faster, cheaper options. One example is Kimi K3 from Moonshot AI. Its team has claimed that Kimi K3 beats GPT-6.1 Sol on coding leaderboards by a meaningful Elo margin.
Codex Agentic Workflows Move Onto Google's Infrastructure
Codex, OpenAI's tool for coding with AI agents, is the other major part of the deal. It could be especially important for software teams. OpenAI and Google Cloud are presenting Codex as more than a tool that suggests lines of code. They describe it as an agent that can plan coding work, carry it out, and check the results within a company's existing Google Cloud environment. This matters because Codex could use Google's data, storage, and computing resources directly instead of requiring information to be moved between different cloud platforms.
For software teams that already use Google Cloud for their data systems, this direct connection removes a common problem with AI coding tools. Companies may otherwise need to move sensitive code and data to another provider's systems or build custom connections between cloud platforms. Developers will also be able to connect local Codex command-line and desktop tools to Google's platform for running AI models. This gives them a way to run coding sessions using OpenAI models hosted on Google Cloud without leaving the tools they already use.
How Enterprises Will Buy the Models
For large companies, the way they buy the models is almost as important as how the technology works. Google Cloud confirmed that customers will be able to buy OpenAI models through the Google Cloud Marketplace. Businesses already use this online store to buy and pay for many software products that run on Google's infrastructure. Importantly, companies with existing spending commitments to Google Cloud will be able to use those commitments to pay for OpenAI models. They will not need a separate budget or a new vendor relationship if they already have an agreement to spend money on Google's cloud services.
This could make buying the models much easier. Large companies often commit to spending money on cloud services over several years. Moving some of that money to a new supplier can be difficult and may require renegotiating contracts. Allowing existing Google Cloud commitments to cover OpenAI model use reduces that difficulty. Companies already paying for Google Cloud can try OpenAI's models without starting a completely new purchasing process.
Google Antigravity and the Gemini Agent Also Get OpenAI Options
OpenAI's models are also planned for Google Antigravity and the Gemini Agent, two Google tools for developers and for building AI agents. This is important because Google is not just offering OpenAI's models as a separate product. It plans to connect them to tools its own developers already use. Developers would then be able to choose which AI model powers their work without leaving Google's platform.
Once this feature becomes available, developers using Gemini Agent or Antigravity should be able to choose GPT-6 Astra, Sol, or Luna as the model that reasons through an agent's tasks. They would still be able to use Google's data connections and deployment tools. However, neither company has announced a firm date for this specific connection. It is a stated plan, not a feature confirmed as available at the time of the announcement.
The Only Hyperscaler With All Three Leading Closed-Source LLMs
Google Cloud is focusing heavily on the number of models customers can choose from. The company says it is now the only large cloud provider offering access to all three leading groups of closed-source large language models, as well as more than 200 other models in Model Garden. This is a notable strategy because Google is using a competitor's product to help persuade customers to choose Google's own cloud services over Microsoft's or Amazon's.
The strategy also shows that Google believes offering a wide choice of models may be more important than having exclusive access to one. Microsoft Azure gained an early advantage in business AI by offering special early access to OpenAI's models. AWS Bedrock built its offering around a wide range of models from other companies, including Anthropic's Claude. Google's move suggests it believes cloud providers will compete by offering their own models, such as Gemini, alongside as many outside models as possible. Businesses can then choose the best model for each job instead of committing to one model through a single contract.
Competitive Landscape: OpenAI, Google, Microsoft, Amazon, and Anthropic
This deal is part of a wider competition. Before the announcement, OpenAI's models were already available through Microsoft Azure and AWS Bedrock. Adding Google Cloud gives OpenAI a third major cloud platform for distributing its models rather than its first. OpenAI's models will now be offered through all three leading US cloud providers. No other major AI model maker, including Google with Gemini and Anthropic with Claude, currently has the same level of access across all three providers, particularly on Google's own platform.
| Cloud Platform | Own First-Party Models | OpenAI Models Available | Notable Third-Party Models |
|---|---|---|---|
| Microsoft Azure | None (relies on OpenAI partnership) | Yes, longstanding integration | Various via Azure AI Foundry |
| AWS Bedrock | Amazon Nova family | Yes, already available | Anthropic Claude, Meta Llama, others |
| Google Cloud | Gemini family | Yes, newly announced (private preview) | 200+ models in Model Garden |
Anthropic is still the clearest example of a major AI company that has not followed this same distribution pattern. Claude is widely available through AWS Bedrock and Google Cloud, but Anthropic has kept its business arrangements with the two cloud providers separate. Nothing in this week's announcement changes that. Readers interested in how Claude's prices compare with OpenAI and Gemini can read Shattered.io's earlier analysis of the price differences between Claude, ChatGPT, and Gemini. Depending on the model tier and how it is used, the difference can reach 20 times.
From Exclusive Azure Deal to Multi-Cloud Distribution
Just a few years ago, the situation was very different. OpenAI's business relationship with Microsoft was close to exclusive. For years, Azure was the only major cloud platform where businesses could license OpenAI's models with full enterprise support. This arrangement helped Microsoft gain an early lead in business AI sales. Microsoft offered OpenAI's models through products such as Microsoft 365 Copilot and Azure AI Foundry, long before competitors had similar options.
AWS Bedrock later began offering OpenAI models, weakening that exclusive arrangement. This week's Google Cloud announcement now extends OpenAI's distribution across all three leading US cloud providers. It also reflects a wider change in the AI industry in 2026. AI companies increasingly treat distribution as separate from what makes their models different. They can sell the same models through many channels and compete instead on price, the ability to complete multi-step tasks, and tools for business customers, rather than relying on exclusive cloud deals.
This change is not limited to OpenAI. DeepSeek has been moving its own systems for running AI models toward wider distribution this fall. Nvidia has also released open-weight models through its Nemotron family. Both developments put pressure on AI companies that still depend on exclusive deals with a single cloud provider. In this setting, Google's decision to host a competitor's leading models looks less like a special favor to OpenAI and more like a response to what business customers increasingly expect: the ability to move between models should be a standard part of a serious cloud platform, not an expensive extra.
Market Impact: What This Means for Enterprise AI Buyers
For purchasing and engineering leaders, the main benefit is more choice without having to move their systems. A company that already uses Google Cloud for its data warehouse, identity management, and compliance tools will no longer need to consider setting up a separate Azure or AWS account just to access OpenAI's most advanced models. This makes it easier and cheaper to test a specific model. Over time, businesses may be more likely to compare several models directly instead of automatically using whichever provider they already depend on.
The deal could also increase price competition. OpenAI's models, Google's Gemini models, and more than 200 other options will sit alongside one another in the same marketplace. This gives business customers a simpler way to compare the cost of processing tokens and the time models take to respond. When customers can make these comparisons easily, prices often come under pressure. This fits the wider pattern in 2026, when leading AI companies have cut prices sharply. One example is OpenAI's roughly 50% price reduction for the Sol model line compared with its predecessor.
Another possible effect is more use of automatic model routing. Large software teams increasingly build simple systems that send each task to the model that can handle it best at the lowest cost, instead of building everything around one provider. Having GPT-6 Astra, Sol, and Luna in the same Google Cloud account as Gemini makes these systems easier to build and run. Billing, logs, and access controls can all be managed through one cloud provider instead of being spread across two or three.
Governance and Regulated-Industry Implications
The announcement highlights Google's data-management controls and infrastructure as important benefits for businesses and industries with strict rules. This matters in areas such as banking, healthcare, and government. Organizations in these fields often cannot adopt a new AI model until they can run it within an existing environment that meets their security and compliance requirements and has been checked through audits. By offering OpenAI's models within Google's own environment and management controls, rather than requiring data to leave Google's infrastructure, the deal gives regulated customers a way to use GPT-6 Astra, Sol, or Luna without having to start a completely new risk review of a third-party provider.
This may be the deal's most important practical benefit for large companies that are careful about risk, even more than the model names themselves. A bank or hospital system that has spent months getting approval to use Google Cloud will not want to repeat the entire process for every new AI model it wants to test. Adding OpenAI's models to an environment that has already been approved removes one of the biggest delays in adopting AI across large organizations.
What's Still Unconfirmed
Several details that business customers will need before deciding how much to spend have not yet been announced. Neither OpenAI nor Google Cloud has published exact prices for the three models through Model Garden. They have also not provided a specific date for general availability beyond saying it will happen in the coming weeks. The companies have not named any business customers taking part in the private preview, and they have not shared expected usage numbers or the value of the partnership's contracts.
This lack of detail is normal for an announcement made on the same day as a deal of this size. However, it means the commercial information needed to judge whether this is a small change in how models are sold or a major change in how businesses buy AI is still missing. Price lists, customer examples, and a clearer launch date are expected to appear over the following weeks as the private preview expands.
Predictions: Where This Multi-Cloud AI Race Goes Next
- GPT-6 Astra, Sol, and Luna will likely become generally available on Google Cloud within the announced period of “coming weeks.” The launch may be timed to a major Google Cloud or OpenAI developer event rather than happening quietly.
- Microsoft and Amazon are likely to respond within the next quarter by adding more models or changing prices on Azure AI Foundry and AWS Bedrock. This would continue the pattern of competitors matching one another rather than relying on exclusive deals.
- Anthropic will likely remain the major AI company most careful about how its models are offered across different clouds. It may continue to negotiate Claude's availability separately with each platform rather than copying OpenAI's broad distribution strategy.
- Early adoption of OpenAI models on Google Cloud will likely be most visible in industries with strict rules, especially banking and healthcare. The deal directly emphasizes the infrastructure and management controls needed for work that must meet demanding compliance requirements.
- Prices for OpenAI models on Google Cloud will likely stay close to OpenAI's own API prices rather than include a special Google discount. This would be consistent with the way AWS Bedrock has priced third-party models so far.
Frequently Asked Questions
What did OpenAI and Google Cloud actually announce on October 21, 2026?
OpenAI and Google Cloud announced a partnership to bring OpenAI's advanced models, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, to Google Cloud's Model Garden. The deal also brings OpenAI's Codex tools for completing multi-step coding tasks to the platform.
Can I use OpenAI's models on Google Cloud right now?
Access is currently limited to a private preview for selected business customers. General availability is planned for the coming weeks, but the companies have not announced an exact date.
How will enterprises pay for OpenAI's models on Google Cloud?
Businesses will buy the models through the Google Cloud Marketplace. They can use existing commitments to spend money on Google Cloud to pay for OpenAI model usage.
What's the difference between GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna?
Astra is the most powerful option and is designed for complex coding, computer use, and scientific research. Sol offers similar intelligence at a lower cost and is aimed at coding agents and professional work. Luna is faster and cheaper, making it suitable for everyday tasks that need to be handled at high volume.
Was OpenAI already available on other cloud platforms before this deal?
Yes. OpenAI's models were already available through Microsoft Azure and AWS Bedrock. Google Cloud is now the third major US cloud provider to offer them.
Will OpenAI's models work inside Google Antigravity or the Gemini Agent?
Both companies say they plan to make OpenAI models available in Google Antigravity and the Gemini Agent. However, neither company has announced a firm date for this connection.
Does this mean Google Cloud now offers every major AI model?
Google Cloud says it is the only large cloud provider offering all three leading groups of closed-source large language models, alongside more than 200 other models in Model Garden. However, the description of these models as “leading” comes from Google and is not an independently verified count.
Why would Google Cloud sell a competitor's AI models?
Google Cloud is making a wide choice of models, rather than exclusive access to one model, a key reason for customers to choose its platform over Microsoft Azure or AWS Bedrock. It is betting that businesses will prefer a cloud service that offers many models alongside Google's own Gemini family.