Claude on Google Cloud (Anthropic-Managed): Setup & Pricing

Claude on Google Cloud now comes in two forms. The new one, Claude Models on Google Cloud (Anthropic managed), went generally available on October 22, 2026. @claudeai announced it as "the full set of Claude Platform features on Google Cloud, with Google Cloud IAM credentials, billing, and commitment retirement." Google's own launch post says both options ship the same models on day zero, at the same pricing, on the same Google Cloud bill.

This guide covers what changed, how to set up Claude on Google Cloud step by step, what it costs, and how to choose between the two options. This page is kept by a Sumus AI Employee and updated as Anthropic and Google change the docs.

What Anthropic-managed Claude on Google Cloud is

Until now, Claude on Google Cloud meant Model Garden on Vertex AI (now Gemini Enterprise Agent Platform): Google runs inference, and Claude Platform features arrive once Google integrates them. The Anthropic-managed option flips that. Anthropic operates the stack as data processor under its Commercial Terms, and Google Cloud handles identity, audit logging and billing.

• Same API as api.anthropic.com. Request bodies, streaming, anthropic-version and anthropic-beta headers are unchanged. Only the base URL and auth differ.
• Day-zero features. Managed Agents, Agent Skills, Files API, code execution, web search and fetch, the memory and advisor tools, the MCP connector, batches and Opus 5.5 fast mode.
• No API keys. Every call carries a Google OAuth access token. ANTHROPIC_API_KEY is ignored.
• A new Anthropic organization. Setup creates a separate org per Google Cloud project, with its own Claude Console. Keys, workspaces, ZDR and settings from an existing Anthropic or Claude Enterprise org don't carry over.
• US and Global endpoints at launch. Location stays global; set inference_geo to "us" per request for US-only inference.

How to set up Claude on Google Cloud (Anthropic managed)

Anthropic's setup docs list four steps. You need Marketplace purchase rights on the billing account and permission to enable services on the project.

1. Subscribe in Marketplace. Open Google Cloud Marketplace, find "Claude Models on Google Cloud (Anthropic managed)", choose Subscribe, then Manage on Console. Existing discount? Talk to your account rep first; discounts can't be applied to past usage.
2. Enable two services. Run: gcloud services enable claude.googleapis.com anthropic-managed-claude.cloudpartnerservices.goog --project=my-project-id
3. Activate the Anthropic organization. On the Claude page choose Activate, enter a contact email, open the setup link, fill in organization name, entity type, country and intended use, and accept Anthropic's Commercial Terms and Usage Policy.
4. Open the Claude Console and copy the workspace ID. It starts with wrkspc_. Developers need it for every call.

IAM roles

Your first call

Install the SDK with pip install -U "anthropic[google_cloud]", run gcloud auth application-default login, then set ANTHROPIC_GOOGLE_CLOUD_PROJECT and ANTHROPIC_GOOGLE_CLOUD_WORKSPACE_ID. In Python, create AnthropicGoogleCloud() and call client.messages.create(model="claude-opus-5-5", ...) exactly as you would on the Claude API. For raw REST, the base URL is https://claude.googleapis.com/v1/projects/{project}/locations/global/workspaces/{workspace}/invoke, with an Authorization: Bearer token and no x-api-key or anthropic-workspace-id header.

Common first-call errors: a 403 means claude.googleapis.com isn't enabled or the caller lacks roles/claude.user; "Model is not available for this organization's inference configuration" means that model isn't enabled for your new org; an empty Usage page just needs a few minutes.

Claude on Google Cloud pricing

Claude on Google Cloud costs the same as the Claude API. Usage is billed through Google Cloud Marketplace in CCUs and draws down eligible Google Cloud commitments. List prices from Anthropic's pricing page:

Opus 5.5 fast mode, up to 2.5x faster output, is $8 input and $40 output per million tokens. Anthropic lists it as working on the Anthropic-managed option; it is not offered on the Google-managed one. Private offers and negotiated discounts go through Marketplace.

Anthropic-managed vs Google-managed: which to choose

• Choose Anthropic-managed if you want new Claude models and betas the day they ship, Managed Agents, skills and code execution, fast mode, and the Claude Console, while keeping one Google Cloud invoice.
• Choose Google-managed if compliance needs Google as sole data processor: VPC Service Controls, Customer-Managed Encryption Keys, US/EU/APAC data residency, FedRAMP High, IL4/IL5, HIPAA eligibility, or native BigQuery, Model Armor and Agent Engine integrations.
• Moving from an existing Anthropic org? Plan a cutover: files, skills, batches and Managed Agents resources stay in the old org, and access is re-granted through IAM.

Managed Agents sessions and scheduled deployments keep running after you revoke the creator's IAM access, so review them when you rotate service accounts.

Skip the cloud setup: use Claude Opus 5.5 in Sumus

Claude on Google Cloud makes sense when your team already lives in GCP and needs IAM, audit logs and commitment drawdown. If you just want to work with Claude, the Marketplace subscription, two services, a new Anthropic org, IAM roles and workspace IDs are a lot of plumbing. In Sumus you sign up and use Claude Opus 5.5 (the default), Sonnet 5.5, Fable 5.1 and Haiku 5.5 directly, with 1M context and 128K output, no separate Claude account and no API key.

AI teammates in Sumus take a task in plain words, run on Claude, and can bring in GPT-6, Grok or Gemini to review the work. See the Claude in Sumus page.

What we're watching

• More regions beyond US and Global for the Anthropic-managed endpoint.
• Features from the Anthropic-managed list reaching the Google-managed option.
• How quickly the next Claude model lands on Claude on Google Cloud.