AI Industry

Anthropic-Managed Claude Goes GA on Google Cloud

Oct 23, 2026·5 min read·By Rahish Saifi

Two people in dark suits, seen from behind, pass a printed invoice and a contract folder with a pen across a walnut conference table in front of a city skyline at dusk.

On October 23, 2026, Google Cloud and Anthropic made a second, Anthropic-operated deployment option generally available: you can now buy and run frontier Claude models through Google Cloud under that Anthropic-managed option, alongside the existing Google-managed path. In a Google Cloud blog post, Michael Gerstenhaber, VP of Product Management for Gemini at Google Cloud, framed both routes as delivering the same frontier models on day zero, the same pricing, and the same Google Cloud billing. The fork is not about model quality or list price—it is about who acts as data processor and whether you receive new Claude Platform capabilities at launch or after Google integrates them.

Google Cloud had already signaled broader partner-model distribution—OpenAI's frontier models are headed to Model Garden—and Anthropic's stack is now purchasable in two governance wrappers on one invoice.

What launched

Google Cloud's announcement makes the Anthropic-managed deployment option for Claude models generally available. Customers could already consume Claude through Google's first-party managed service; GA adds a second, explicitly Anthropic-operated lane that still bills through Google Cloud.

Both paths expose the same named model family, per Google Cloud and Anthropic: Claude Fable 5.1, Claude Sonnet 5.5, Claude Haiku 5.5, and Claude Opus 5.5. Anthropic's companion article states that the full set of Claude API features, plus new features and betas, ship on Google Cloud the same day they appear on the native Claude Platform when you choose Anthropic-managed—examples Google Cloud cites include Claude Managed Agents (beta), skills, and code execution.

Google Cloud's announcement says Anthropic operates the stack as data processor on Google Cloud infrastructure under Anthropic's terms, but Google Cloud's setup documentation adds that data storage and inference occur on Anthropic-managed infrastructure that may route outside Google Cloud, with an inference_geo parameter to pin inference to a geography. Customers still authenticate with Google Cloud IAM, retain Cloud Audit Logging per Google Cloud and Anthropic, and consolidate spend on Google Cloud billing, including drawdown against eligible commitments, as Anthropic's article describes. Endpoints are US and Global today in Anthropic's public guidance.

Anthropic-managed vs Google-managed

Dimension	Anthropic-managed	Google-managed
Data processor	Anthropic operates the stack as processor on GCP; terms per Anthropic (Google Cloud blog)	Google Cloud is processor under existing customer agreements (Google Cloud blog)
Platform features at launch	Day-zero Claude Platform features—Managed Agents (beta), skills, code execution (Google Cloud blog; Anthropic article)	Same frontier models on day zero; new platform features roll out over time once integrated (Google Cloud blog)
Models & pricing	Same Fable 5.1, Sonnet 5.5, Haiku 5.5, Opus 5.5; same pricing (Google Cloud blog)	Same models and pricing (Google Cloud blog)
Billing & auth	Google Cloud billing, IAM, audit logs (Google Cloud and Anthropic), commitment drawdown (Anthropic article; Google Cloud docs)	Same billing surface; Google-native controls (Google Cloud blog)
Compliance & residency	Anthropic processor model; see Anthropic terms and setup docs	VPC Service Controls, CMEK, data residency in US, EU, and APAC; integrations such as BigQuery and Model Armor (Google Cloud blog)
API surface (docs)	Claude API endpoints (`/v1/{endpoint}`), model in body, `anthropic-version` header, `stream:true`; location global; Marketplace billing; rate limits by Anthropic (Claude docs)	Agent Platform prediction endpoints (model in URL, separate `:streamRawPredict`); rate limits by Google Cloud (Claude docs)
Best when	You want Claude Platform parity and Anthropic as processor while staying on one GCP invoice (Anthropic article)	You need Google as processor, strict regional residency, or Model Armor / Agent Engine (Anthropic article; Google Cloud blog)

What developers change in code

The split is not cosmetic—it changes URLs, request shape, and who throttles you. Anthropic's Claude on Google Cloud overview documents two integration styles.

Under Anthropic-managed, you call familiar Claude API endpoints: /v1/{endpoint} with the same request format as the Claude API, the model specified in the request body, the anthropic-version header, and streaming via stream:true. Billing flows through the Google Cloud Marketplace; rate limits are managed by Anthropic. Setup provisions a new Anthropic organization for your Google Cloud project—more on that below.

Under Google-managed, you use Agent Platform prediction endpoints instead: the model appears in the URL, and streaming uses a dedicated :streamRawPredict path. Google Cloud manages rate limits. That path aligns with teams already wiring Claude through Google's agent and security tooling rather than lifting Anthropic SDK calls verbatim.

Setup, billing, and the new-org gotcha

Google Cloud's Anthropic-managed setup guide walks through a Marketplace purchase flow: choose Anthropic managed on the Claude models page, enable the Claude API, buy via Marketplace at list price or through a negotiated discount or private offer, then activate Claude Console, which provisions a new Anthropic organization for the project.

That provisioning step is the operational surprise. Google Cloud's setup documentation notes that API keys, workspaces, HIPAA controls, Zero Data Retention settings, and other configuration from an existing first-party Anthropic org do not carry over to the org created for Anthropic-managed on GCP. It also says Google Cloud and Anthropic act as separate, independent data processors for their respective parts of the service—relevant for legal review when you already run Claude natively and are adding a GCP org.

On discounts, Anthropic's article is explicit: customers with existing discounts should contact their Anthropic or Google Cloud account executive before starting, because discounts cannot be applied retroactively. That is easy to miss when Marketplace checkout feels self-serve.

Who should pick which path

Choose Anthropic-managed if your priority is feature parity with Claude Platform on day zero—agents, skills, code execution, and future betas—while keeping Google IAM, audit logs, and commitment drawdown. Anthropic positions this path for teams that are comfortable with Anthropic as processor but want a single Google Cloud invoice.

Stay on Google-managed if procurement requires Google Cloud as data processor under your existing agreement, if you need VPC Service Controls, CMEK, or data residency in the US, EU, or APAC regions Google highlights, or if you rely on native hooks such as BigQuery, Model Armor, or Agent Engine. Anthropic's guidance also steers residency- and safety-sensitive buyers toward Google-managed when Model Armor or Agent Engine is non-negotiable.

Neither option replaces architectural diligence: model names and list pricing may match, but processor identity, API shapes, rate-limit owners, and feature velocity do not.

What it means for the market

For Google Cloud, GA Anthropic-managed Claude strengthens its pitch of offering more than one frontier lab's models on a single bill, alongside OpenAI's upcoming Model Garden presence. For Anthropic, it exports Platform velocity—agents, skills, execution—into enterprises that will not open a second procurement lane for every beta.

For builders, the practical lesson is to decide early based on governance versus feature velocity, then lock API integration style to that decision. Swapping later may be possible, but org migration, discount timing, and separate processor boundaries make mid-flight changes expensive.