Google Cloud makes an Anthropic-managed Claude option generally available

What Google announced

Google Cloud said on 22 October 2026 that an Anthropic-managed deployment option for Claude is now generally available. Until now, companies that bought Claude through Google Cloud got a version that Google itself runs and governs. The new option lets Anthropic run the service instead, still on Google Cloud infrastructure and still paid for through a Google Cloud bill.

The announcement came in a short post from Google Cloud's vice president of product management for Gemini. It frames the change as a matter of choice: one Claude, two ways to deploy it, and a customer picks based on who they want handling their data and how quickly they want new Claude features.

It lands one day after OpenAI said its frontier models and Codex are coming to Google Cloud, so Google is now leaning hard into selling rival labs' models next to its own Gemini.

Two ways to run the same Claude models

Google lists the two options as Claude Models on Google Cloud (Anthropic managed), which is the new one, and Claude Models on Google Cloud (Google managed), which is the existing service. According to the post, both deliver the same frontier models on the day Anthropic releases them, at the same prices, on the same Google Cloud bill.

The differences are in two places. The first is who acts as the data processor, the company legally responsible for handling the prompts and outputs a customer sends. The second is how fast customers get new Claude Platform features, the tools Anthropic builds around its models for developers, such as agents that run on Anthropic's servers.

Google publishes a comparison table. Model availability is day zero for both. First-party tools and betas are day zero on the Anthropic-managed option, and arrive on the Google-managed option only once they are integrated into Google's infrastructure. Pricing is standard on both and counts toward eligible Google Cloud commitments.

Who processes your data

On the Anthropic-managed option, Anthropic operates the stack as the data processor on Google Cloud infrastructure, under Anthropic's Commercial Terms of service. That means the contract governing how Claude requests are handled is Anthropic's, not Google's, even though the servers sit in Google data centres.

On the Google-managed option, Google Cloud stays the data processor under a customer's existing Google Cloud agreements. For a bank, hospital or government buyer that has already negotiated data terms with Google, that is a meaningful difference. Adding Anthropic as a processor may mean a new vendor review, a new data protection agreement and fresh questions from compliance teams.

For smaller teams that mostly want the newest Claude tools without opening a separate Anthropic account and separate invoice, the Anthropic-managed route may be the simpler one. Google's post does not say whether data sent to the Anthropic-managed option can be used to train models; that would be set by Anthropic's Commercial Terms, which customers should read before switching.

Which features arrive first, and where

Google names three Claude Platform features that customers get on day zero with the Anthropic-managed option: Claude Managed Agents, which is still in beta, skills, and code execution. Managed Agents is Anthropic's service for running agents on its own servers; this month it gained dynamic workflows that let one agent orchestrate many others and OpenTelemetry tracing for agent runs.

Features like these have so far been easiest to reach through Anthropic's own API. The new option is a way for Google Cloud customers to get them without leaving Google's identity, logging and billing systems. On the Google-managed option, Google says new Claude Platform features roll out over time as they are integrated into its infrastructure. It gives no timeline and does not say which features are planned.

The post's get-started link points to a Claude Platform page in the Google Cloud console, under the Agent Platform section. Google's Claude Model Garden page now refers to Vertex AI as Agent Platform, which is the product family these Claude listings belong to.

Security and compliance controls

Both options keep some Google Cloud controls. With the Anthropic-managed version, customers still sign in and grant access with Google Cloud IAM, the identity and access management system, and keep Cloud Audit Logging, which records who did what.

The Google-managed option keeps a longer list. Google says it provides VPC Service Controls, which fence services off inside a private network perimeter, Customer-Managed Encryption Keys (CMEK), and data residency in the US, the EU and the Asia-Pacific region. It also highlights native integrations with BigQuery, Google's data warehouse, and Model Armor, its screening service for prompts and responses.

Google's announcement does not say whether VPC Service Controls, CMEK or regional data residency are available on the Anthropic-managed option. Read plainly, the post lists them only for the Google-managed one, and it describes that option as the choice for strict governance, compliance and sovereignty requirements.

Billing and commitments

Money is the clearest shared feature. Google says both options use standard pricing, are billed through Google Cloud and Google Cloud Marketplace, and draw down eligible Google Cloud commitments. Large companies often sign multi-year spending commitments with a cloud provider, and being able to count AI model spending against that commitment makes the purchase easier to approve.

That is the same pitch OpenAI made a day earlier for its models on Google Cloud. For Anthropic, it means customers can use its newest platform features while their finance team keeps dealing with a single cloud vendor.

The models on offer

Google lists Claude Fable 5.1, Sonnet 5.5, Haiku 5.5 and Opus 5.5 as available whichever option a customer picks. The Model Garden page describes Fable 5.1 as a frontier model for long-running, high-stakes work, and Haiku 5.5 as fast, cost-efficient intelligence for high-volume, latency-sensitive agentic work, subagents and real-time experiences.

The same Model Garden page mentions Provisioned Throughput, which reserves dedicated capacity at a fixed cost, and Google's agent tools such as Memory Bank and Sessions for keeping context across conversations. The announcement itself does not say which of those Google Cloud extras work with the Anthropic-managed option.

Why this matters

Cloud providers have long resold other companies' models. What is new here is a model maker running its own managed service inside a hyperscaler's console, with the cloud provider handling identity and billing. It moves the relationship closer to a marketplace: Google provides the shop, the sign-in and the invoice, and Anthropic runs the product.

For Anthropic, it shortens the gap between launching a feature and having Google Cloud customers use it. For Google, it keeps those customers' spending on Google Cloud even when they pick a rival's model. Claude already reaches Google's productivity apps through Claude for Google Workspace, so the two companies' products are increasingly woven together.

For buyers, the main consequence is a decision that did not exist before. Choosing a Claude deployment on Google Cloud is now partly a legal and compliance choice, not only a technical one.

How to choose, and what is still unclear

Teams that need data residency, customer-held encryption keys or a private network perimeter should, on the evidence of the post, stay with the Google-managed option. Teams that want Managed Agents, skills and code execution on launch day, and are comfortable with Anthropic as data processor, now have a supported way to get them through Google Cloud.

Several things are not yet clear from public material. Google does not list which regions host the Anthropic-managed service, whether its rate limits differ from the Google-managed one, or how support is split between the two companies when something breaks. It also does not say whether Google-managed customers will get every Claude Platform feature eventually, or only some. AI Lookout will update this story if Google or Anthropic publish more detail.