Claude API run by Anthropic is now available through Google Cloud

If your company buys cloud through Google, you can now call the Claude API Anthropic runs, Claude Managed Agents included, and pay for it on your Google Cloud bill. A few features are missing (see below). Our read: plan it as a move to a new Anthropic organization, not a settings change.

What changed#

Anthropic's release notes for Thursday read: "We've launched Claude Models on Google Cloud (Anthropic managed), the Claude API operated by Anthropic and reached through your Google Cloud project." At 5:21 PM ET, @claudeai posted: "Claude Models on Google Cloud (Anthropic managed) is now generally available." Until now, Claude on Google Cloud meant the Google managed offering, which Anthropic's comparison page describes as "Google operates Claude models as part of Google Cloud's Agent Platform." That one stays.

	Google managed (existing)	Anthropic managed (new)
API	Agent Platform prediction endpoints	Claude API endpoints
Data processor	Google Cloud	Anthropic
New platform features	Roll out over time, per Google	Same day as the native Claude Platform, per Anthropic
Location	Global, multi-region or regional	global
Rate limits	Google Cloud	Anthropic

Why it matters#

Our read: Google Cloud teams no longer have to choose between their own billing and Anthropic's newest tools. Anthropic's article says "Billing is through a single Google Cloud invoice and Claude usage draws down your existing Google Cloud commitment." The docs list Claude Managed Agents as available there, including scheduled deployments and credential vaults, plus fast mode for Claude Opus 5.5. Anthropic's article and Google's blog both label Managed Agents a beta.

How to use it#

1. Subscribe to the listing in Google Cloud Marketplace, then click Manage on Console.
2. Enable both services on each calling project: gcloud services enable claude.googleapis.com anthropic-managed-claude.cloudpartnerservices.goog --project=my-project-id.
3. Click Activate, complete the form from the emailed link, and accept Anthropic's terms.
4. Click Open Claude Console and copy the workspace ID, which starts with wrkspc_.
5. Grant roles/claude.user, export ANTHROPIC_GOOGLE_CLOUD_PROJECT and ANTHROPIC_GOOGLE_CLOUD_WORKSPACE_ID, and swap your client (Python: from anthropic import AnthropicGoogleCloud). Per the docs, auth moves to a Google access token sent as Authorization: Bearer instead of x-api-key; request bodies, streaming and responses stay the same.

A read-only audit prompt for Claude Code (ours, untested):

Read this repo and list every place it creates an Anthropic client, sends an x-api-key or anthropic-workspace-id header, uses MCP tunnels, or uses a model ID with an @ before a date. For each, say what would change on Claude Models on Google Cloud (Anthropic managed), using only the docs page I paste below. Don't edit any files.

Good to know#

• It's a new organization. "These steps create a new Anthropic organization for your Google Cloud project." "API keys, workspaces, and Claude Console settings don't carry over." "Files, skills, batches, and Claude Managed Agents resources stay in the existing organization." (docs)

• No API keys, no MCP tunnels. "Authentication is through Google Cloud credentials only." "MCP tunnels aren't available on Claude Models on Google Cloud (Anthropic managed)." (docs)

• Revoking access doesn't stop running sessions. "If you revoke Google Cloud access from the identity that created a session, the session doesn't stop." (docs) The same goes for scheduled deployments: "A deployment doesn't re-check its creator's access on each scheduled run. To stop it, pause, update, or archive it." (docs)

• ZDR doesn't follow you. A zero data retention arrangement on another Anthropic organization "doesn't carry over" (docs).

• Turn on audit logs. Data Access logs cover most routes, including messages, and the docs say: "They're off by default." The same page tells you to turn them on for claude.googleapis.com in your project's audit log settings if you want a record of requests.

• Our read: ask about discounts first. "Discounts can't be applied to past usage." (docs)

• The SDK client is beta, per the docs.

• The location wording differs. Anthropic's article says: "Claude Models on Google Cloud (Anthropic managed) is currently available with US and Global endpoints." The docs say: "Leave the location at global. The location is separate from where inference runs". For US-only inference, the docs point you to setting inference_geo to us on each request. Our read: the article's "US" may mean that setting rather than a second location, but Anthropic hasn't said.

What to watch: Our read: whether the docs add a US location or the article's wording changes, and whether the docs start labeling Managed Agents a beta the way the article and Google's blog do.

Last checked: Oct 23, 2026, 6:11 to 6:25 AM ET (release notes, both Claude docs pages, Anthropic's article, Google's blog and both X posts). What changed: first published Oct 23; no updates since.