The full Claude Platform reaches Google Cloud, with a choice of who handles your data

On 22 October 2026, Anthropic made Claude Models on Google Cloud (Anthropic managed) generally available. It is the Claude API, run by Anthropic and reached through a customer's Google Cloud project. Google Cloud IAM, Google's system for who may do what, controls access, and Google Cloud Marketplace handles billing.

Anthropic says customers get the full set of Claude API features. New features and betas ship on the same day they go live on Anthropic's own Claude Platform. The models listed are Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, with new models added as they launch.

The listed features include Claude Managed Agents in beta, for long-running agents without managing infrastructure. Code execution runs Python and analyses data within an API call, and computer use lets Claude operate a browser or desktop. The MCP connector, also in beta, connects Claude to other tools without writing client code.

Agent skills, the advisor strategy in beta, the Files API, citations, web search, web fetch, prompt caching and batch processing are on the list too.

Fast mode on Claude Opus 5.5 is generally available on this option. Anthropic says it generates output up to 2.5 times faster, for live debugging and interactive agents. It costs $8 per million input tokens and $40 per million output tokens.

Customers also get the Claude Console, Anthropic's development environment for agents. Teams use it to write prompts, configure agents and run them on hosted environments. It also shows usage and costs, and sets access and security for each workspace.

Usage goes on a single Google Cloud invoice and draws down existing Google Cloud commitment. Anthropic sends no separate invoice. The option is currently available with US and Global endpoints.

Google's blog post of 23 October adds that the option runs on Google Cloud infrastructure under Anthropic's Commercial Terms of Service. It works with Google Cloud IAM and Cloud Audit Logging, at standard pricing.

For a company with a Google Cloud commitment, Claude usage on this route is paid from money already committed to Google. It arrives on the invoice finance already processes, because Anthropic sends no invoice of its own.

Subscribing to the Marketplace listing means accepting Anthropic's Commercial Terms of Service and Usage Policy, so your legal team still has terms to read.

Anthropic presents feature parity as the main reason to pick this route. Managed Agents, the MCP connector and the advisor strategy are all betas. Anthropic says betas reach this option on the day they reach its own platform.

Fast mode is billed by the token, at $40 per million output tokens. A team that uses it for live debugging should watch the Usage and Cost pages in the Claude Console from the first week.

Claude remains available through the Google managed option, where Google is the data processor. A data processor is the company that handles personal data on the customer's behalf. Google's blog says both options have the same pricing and billing, including drawdown of eligible commitments.

On Anthropic managed, Anthropic operates the service and is the data processor. Anthropic's docs say some supporting services, such as its trust and safety systems, process data outside Google Cloud. Data retention follows the Claude API's policy, and Zero Data Retention is available on request through an Anthropic account representative.

On Google managed, Google says new Claude models arrive on day zero, and new platform features roll out over time as they are integrated. Google is the data processor under existing customer agreements.

Google lists VPC Service Controls, its network perimeter controls, and Customer-Managed Encryption Keys (CMEK) for its option. It also lists data residency in the US, the EU and Asia Pacific, and native links to BigQuery and Model Armor.

In its article, Anthropic says its own option suits customers who want day-zero access to every beta and generally available feature. Google managed remains the better fit for customers who need Google Cloud as the data processor or have strict regional residency rules. It also suits those who want Google tools such as Model Armor or Agent Engine.

Google managed offers data residency in the US, the EU and Asia Pacific. Anthropic managed lists US and Global endpoints only, and no source names a UK or Gulf location for either option. An EU bank or insurer with residency rules will likely stay on Google managed for now.

Teams that want the newest features, such as Managed Agents or the MCP connector, get them first on Anthropic managed. Their security team then has Anthropic to assess as a data processor, plus some processing outside Google Cloud to account for.

On our reading, a company can use both options at once. They share Google Cloud credentials and the Google Cloud bill, and Anthropic's SDKs support both. Regulated workloads could run on Google managed, and feature-heavy development work on Anthropic managed.

Setup creates a new Anthropic organisation for the Google Cloud project. Anthropic's docs say it is separate from any organisation the company already has with Anthropic, Claude Enterprise included.

API keys, workspaces and Claude Console settings don't carry over. Access is granted through three Google Cloud IAM roles: roles/claude.admin, roles/claude.user and roles/claude.viewer.

The two organisations are independent and can serve traffic in parallel, so a company can move gradually. The existing organisation keeps its current billing. Anthropic manages rate limits on the new one, and advises checking them before shifting production traffic.

A Zero Data Retention arrangement on another Anthropic organisation doesn't carry over to the new one.

Four things aren't supported: API keys, MCP tunnels, memory stores on self-hosted environments, and the pre-built self-hosted sandbox workers in the ant CLI and SDKs. A team that needs a self-hosted worker writes its own against the Environments Work endpoints.

A company with a direct Claude API organisation or Claude Enterprise ends up with two Anthropic organisations. Each has its own access, workspaces, rate limits and billing, so admins set the new one up from scratch.

Zero Data Retention is easy to miss. A company that agreed it on its existing organisation has to request it again before sensitive traffic moves to the new one.

Running both in parallel allows a move one workload at a time. The old organisation keeps billing as it does today, so finance will see Claude costs in two places until the move is done.