Claude on Google Cloud, run by Anthropic: the full API with Google IAM and billing

Until October 22, 2026, a company on Google Cloud had two ways to use Claude. Call Anthropic's API directly, with a separate contract and an Anthropic invoice. Or use Claude through Vertex AI, where Google runs the service and new Anthropic features arrive later, or not at all.

Claude Models on Google Cloud (Anthropic managed) is a third option. It is Anthropic's own Claude API, operated by Anthropic, reached through your Google Cloud project. Google handles who can call it and how you pay. Anthropic runs everything else.

What you get

Every Claude Platform feature, on launch day. That includes Managed Agents (agents, environments, sessions, credential vaults, memory stores, scheduled deployments, self-hosted sandboxes), code execution, computer use, skills, the Files API, citations, web search and fetch, the MCP connector, prompt caching, and batches. New betas ship here the same day they ship on the direct API.

The Claude Console. You get the full Console for prompts, agent configuration, usage and cost pages, and workspaces.

All current models. Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5, with the same model IDs as the direct API. Fast mode for Opus 5.5 is available at $8 per million input tokens and $40 per million output.

One bill. Usage appears on your Google Cloud invoice and draws down an existing Google Cloud spend commitment. If you have a negotiated Anthropic discount, confirm it with your account team before you start: discounts can't be applied to past usage.

How it differs from Vertex AI

	Anthropic managed	Google managed (Vertex AI)
Who runs it	Anthropic	Google
Data processor	Anthropic	Google
New features	Same day as Claude API	Later, sometimes not at all
Managed Agents	Yes	No
Console	Claude Console	Google Cloud console
Regions	US and Global	Google regions, strict residency
Google-native tools (Model Armor, Agent Engine)	No	Yes

Pick Vertex if you need Google as the data processor, strict regional residency, or Google-native integrations. Pick Anthropic managed if you want feature parity and Google billing.

What's missing

• No API keys. Authentication is Google Cloud credentials only.
• No MCP tunnels.
• No memory stores on self-hosted environments.
• The pre-built self-hosted sandbox workers don't work here. Write your own against the Environments Work endpoints.

Setting it up

Setup creates a new Anthropic organization tied to your Google Cloud project. It's separate from any Anthropic organization or Claude Enterprise org your company already has, and a zero-data-retention agreement on another org does not carry over.

1. Subscribe to the "Claude Models on Google Cloud (Anthropic managed)" listing in Google Cloud Marketplace.
2. Enable the two services:

gcloud services enable claude.googleapis.com \
  anthropic-managed-claude.cloudpartnerservices.goog \
  --project=my-project-id

3. Grant roles on the project. There are three: roles/claude.viewer (read-only), roles/claude.user (send messages, manage files, batches, skills, agents), and roles/claude.admin (plus org-level settings and Console access).

gcloud projects add-iam-policy-binding my-project-id \
  --member="serviceAccount:my-app@my-project-id.iam.gserviceaccount.com" \
  --role="roles/claude.user"

4. Open the Claude Console from the Claude page in the Google Cloud console (only roles/claude.admin can), go to Settings > Workspaces, and copy the workspace ID. It starts with wrkspc_. Every caller needs it.

Calling it

The SDK client is AnthropicGoogleCloud (in beta). It reads Application Default Credentials and builds the URL from your project and workspace.

pip install -U "anthropic[google_cloud]"
export ANTHROPIC_GOOGLE_CLOUD_PROJECT=my-project-id
export ANTHROPIC_GOOGLE_CLOUD_WORKSPACE_ID=wrkspc_...
gcloud auth application-default login

from anthropic import AnthropicGoogleCloud

client = AnthropicGoogleCloud()

message = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello, Claude"}],
)
print(message.content[0].text)

In TypeScript the package is @anthropic-ai/google-cloud-sdk and the class has the same name. Leave the location at global. For US-only inference, set inference_geo: "us" on the request.

For admins

Access is a Google IAM decision, so it fits whatever review process you already run for Google Cloud roles. A few things to plan for:

• Separate teams need separate projects. IAM roles apply to every workspace in a project. If two teams need different access, give each its own project, which means its own Anthropic org.
• Revoking access doesn't stop running agents. A Managed Agents session or scheduled deployment keeps running after you remove its creator's IAM role. When someone leaves or a service account rotates, review the sessions and deployments that identity created.
• Turn on Data Access audit logs. Cloud Audit Logs record activity under claude.googleapis.com, but Data Access logs, which cover messages, files, and most agent routes, are off by default.
• Compliance API and CMEK are available. Toggle the Compliance API in the Console under Settings > Compliance. Customer-managed keys use Google Cloud KMS.

This also arrives on the same day as API version 2026-10-21, and the new SDK versions are what you'll install. Read that first if you have existing Claude code you plan to move.