Simon Carter
Cloud & Infrastructure

Claude is now available directly through Google Cloud, operated by Anthropic

Anthropic launched Claude Models on Google Cloud (Anthropic managed) on 22 October 2026, bringing same-day feature parity and GCP billing.

cloud infrastructure category

On 22 October 2026, Anthropic launched Claude Models on Google Cloud (Anthropic managed), a new way to call the Claude API from inside a Google Cloud project. Anthropic operates the infrastructure. Google Cloud handles the authentication and the bill. That combination is new, and it matters for teams who spend their lives inside GCP.

What this actually is

Before this launch, Google Cloud customers accessed Claude through Vertex AI, where Google acts as the data processor and new Claude features follow Google’s own integration schedule. That option still exists and is not going away. What Anthropic has now added is a second path: an Anthropic-operated deployment that sits inside your Google Cloud project but runs on Anthropic’s infrastructure under Anthropic’s terms.

The practical effect is that you get the full Claude API, including features and betas that ship the same day they go live on Anthropic’s platform, without waiting for a Vertex AI integration cycle to complete.

This follows a similar move in May 2026, when Anthropic launched Claude Platform on AWS, bringing the same Anthropic-operated model to teams working inside AWS.

How access and billing work

Authentication uses Google Cloud IAM. You enable claude.googleapis.com on your project, and your existing IAM permissions and service accounts work from that point on. There is no separate Anthropic account to set up and no new credential system to manage.

Billing flows through the Google Cloud Marketplace. Anthropic meters token usage in Claude Consumption Units (one unit equals $0.01 USD) and reports it to Google Cloud. The charges appear on your standard Google Cloud invoice, and the spend draws down any existing Google Cloud committed use. You do not receive a separate bill from Anthropic.

In code, the switch is minimal. Developers use the AnthropicGoogleCloud client from the Anthropic SDK:

client = AnthropicGoogleCloud()

Model IDs follow the same naming convention as the standard Claude API, so migration from the native platform is straightforward.

One detail worth knowing: setup creates a new Anthropic organisation scoped to your Google Cloud project, separate from any existing Anthropic organisation your company may already have. Keep that in mind if you are managing multiple Claude deployments across teams.

What you get from day one

The headline feature for this launch is same-day parity. Every new Claude API feature and beta ships to Claude Models on Google Cloud (Anthropic managed) at the same time it goes live on Anthropic’s platform. That includes:

  • Code execution (Python execution, chart generation, data analysis)
  • Agent Skills API (reusable instructions and workflows)
  • Claude Managed Agents beta (hosted infrastructure for long-running agent workloads)

Models available at launch are Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5. New models will ship to this offering as Anthropic releases them.

Fast mode for Claude Opus 5.5 is also available from day one. Fast mode prioritises speed, offering up to 2.5x faster output token generation. Pricing for Opus 5.5 in fast mode is $8 per million input tokens and $40 per million output tokens. That same fast mode pricing applies consistently across Anthropic’s first-party platform, Claude Models on Google Cloud (Anthropic managed), and Claude in Microsoft Foundry on Azure.

Regional availability at launch covers US and Global endpoints.

When to choose this over Vertex AI

Anthropic and Google have both been clear that this is an addition, not a replacement. There are good reasons to stay with the Google-managed Vertex AI option:

  • Your organisation requires Google Cloud as the data processor
  • You have strict regional data residency requirements (Google-managed supports US, EU, and APAC)
  • You need native Google Cloud integrations such as Model Armor or Agent Engine
  • Your existing contracts and compliance posture are built around Google’s terms

If any of those conditions apply, the Google-managed offering remains the right choice.

The Anthropic-managed option makes sense when your team wants immediate access to new Claude capabilities without waiting for feature rollout through Vertex, and when consolidating everything onto the Google Cloud invoice is more important than keeping Google as the data processor.

What this means for your team

If your organisation is already paying for Google Cloud, you can now access the latest Claude features, models, and betas without opening a separate Anthropic contract, provisioning new credentials, or splitting your spending across two invoices. The service plugs into IAM permissions and audit logging your team already manages, and eligible committed Google Cloud spend covers the token usage.

For teams building on Claude Opus 5.5 specifically, fast mode being available at launch rather than as a future addition removes a common source of friction: building on a capability that exists in one environment but not the one you actually deploy to.

The broader pattern here is worth noting. Anthropic is building distribution into the cloud providers’ own billing and identity infrastructure. For enterprise teams, that reduces the procurement and credential overhead that often slows AI adoption. The feature set is the same; the friction to get there is lower.