Anthropic and Google Cloud made a second way to run Claude on Google Cloud generally available on October 22, 2026. In Claude Models on Google Cloud (Anthropic managed), Anthropic operates the Claude API itself on Google Cloud infrastructure, while access goes through Google Cloud IAM and the bill goes through Google Cloud.
The pitch, in Anthropic’s announcement, is “the full set of Claude Platform features” for Google Cloud customers. Claude’s X post went up at 21:21 UTC, and Google Cloud’s post the same day frames it as a choice between two operators.
On this page · 8 sectionsOpen
Claude Models on Google Cloud (Anthropic managed) became generally available on October 22, 2026. Anthropic runs the Claude API on Google Cloud infrastructure.
Google Cloud customers get the full Claude Platform the day features ship, including Managed Agents, code execution, skills and fast mode for Opus 5.5.
Access is Google Cloud IAM only, with no API keys, and usage bills through Google Cloud Marketplace against existing commitments.
Anthropic is the data processor here. The Google-managed option keeps Google as processor, with data residency in the US, EU and APAC.
Fast mode on Opus 5.5 costs $8 per million input tokens and $40 per million output, double the standard $4 and $20, for up to 2.5x faster output.
§ 01What Anthropic launched
Until now, Claude on Google Cloud meant Google running the models inside its Agent Platform, with Claude Platform features arriving once Google integrated them. The new option puts Anthropic in charge of the service. Google’s post puts the shared ground plainly: “Both options deliver the same frontier models on day zero, the same pricing, and the same Google Cloud billing.”
What differs is who operates the stack, who is the data processor, and how fast new features show up.
| Aspect | Anthropic managed (new) | Google managed |
|---|---|---|
| Who operates it | Anthropic, under Anthropic’s Commercial Terms | Google Cloud, under existing Google Cloud agreements |
| Data processor | Anthropic | Google Cloud |
| New Claude features and betas | Same day as the Claude API | Once integrated by Google |
| API | Claude API endpoints at claude.googleapis.com | Agent Platform prediction endpoints |
| Location | Global, with US-only inference per request | Global, multi-region or regional; residency in US, EU, APAC |
| Rate limits | Managed by Anthropic | Managed by Google Cloud |
| Billing | Google Cloud Marketplace, draws down eligible commitments | Google Cloud, eligible commitment drawdown |
| Native Google integrations | Not listed | BigQuery, Model Armor, Agent Engine, VPC Service Controls |
Both run Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 today. Anthropic says new models ship on its managed option as they launch.
§ 02What you get on the Anthropic-managed option
Anthropic’s feature list for the new option reads like its own platform docs, because it is the same API:
- Agents: Claude Managed Agents (beta), including environments, sessions, credential vaults, memory stores, scheduled deployments and self-hosted sandboxes.
- Tools: code execution, computer use, web search and fetch, the memory tool, the advisor tool and the MCP connector (beta).
- Building blocks: Agent Skills, the Files API, citations, prompt caching, structured outputs and the Message Batches API.
- The Claude Console, opened from the Google Cloud console rather than by email sign-in.
- Customer-managed encryption keys through Google Cloud KMS, set up in the Claude Console.
Fast mode for Opus 5.5 is generally available there too, with “up to 2.5x faster output token generation” at a premium.
| Mode | Input | Output |
|---|---|---|
| Standard | $4 | $20 |
| Fast mode | $8 | $40 |
Fast mode doubles both prices. It is not available with the Batch API.
§ 03How access and setup work
Setup runs through Google Cloud Marketplace: subscribe to the listing, enable two services on each project, and finish the organization form. Setup creates a new Anthropic organization for that project, with one Default Workspace.
- No API keys. Anthropic’s docs say it directly: “Authentication is through Google Cloud credentials only.” The Claude Console cannot create keys for the organization setup creates.
- Three IAM roles, granted on the project: Claude Viewer, Claude User and Claude Admin. Only Claude Admin can open the Claude Console.
- One organization per project. Roles apply to every workspace in the project, so teams that need separate access get separate Google Cloud projects.
- Existing discounts: Anthropic asks customers with a discount to contact their account executive first, because discounts cannot be applied to past usage.
§ 04The access detail to read twice
Managed Agents keep running after the person who started them loses access. Anthropic’s docs say “Access is checked when a session or a deployment is created, not while it runs.” A revoked identity’s running sessions and scheduled deployments carry on until someone interrupts, pauses or archives them.
Anthropic’s own advice: when you remove a user or rotate a service account, review the sessions and deployments that identity created. For any team giving agents long-running jobs, that belongs on the offboarding checklist.
§ 05What is not supported
- API keys.
- MCP tunnels.
- Memory stores for Managed Agents sessions on self-hosted environments.
Anthropic notes its list does not cover every Claude API feature, so an unlisted feature is not a promise either way.
§ 06Which option to pick
- Pick Anthropic managed if you want new Claude features and betas the day they ship, Managed Agents, or fast mode, and your Google Cloud commitments should cover the spend.
- Stay on Google managed if Google must be your data processor, you need EU or APAC residency, or you depend on Model Armor, Agent Engine or VPC Service Controls.
- Calling api.anthropic.com with a key today? You use neither option. Anthropic publishes a path for moving those calls onto the Anthropic-managed option.
CellCog’s AI employees run on Claude Opus 5.5. We had no part in this launch.
§ 07Update log
- October 23, 2026: page opened from Anthropic’s and Google Cloud’s announcements and Anthropic’s setup docs. Same day: added an explainer scene of the two options.
§ 08Sources
- Anthropic, Claude Platform is now available on Google Cloud, October 22, 2026.
- Google Cloud, Expanding Claude availability and deployment choice on Google Cloud, October 22, 2026.
- Anthropic docs, Claude Models on Google Cloud (Anthropic managed) and the comparison of both options.
- Anthropic, Pricing, fast mode section.
Q1What is Claude Models on Google Cloud (Anthropic managed)?
A way to call the Claude API from a Google Cloud project. Anthropic operates it, Google Cloud IAM controls access, and usage lands on your Google Cloud bill.
Q2How is it different from the Google-managed option?
On the Google-managed option, Google runs Claude inside Agent Platform and acts as data processor, with regional residency and native Google Cloud integrations. New Claude features arrive there once integrated.
Q3Can I use an Anthropic API key with it?
No. Authentication is through Google Cloud credentials only, and the Claude Console cannot create API keys for the organization that setup creates.
Q4Does it cost more than the Claude API?
Google Cloud says both of its Claude options carry the same pricing. Anthropic lists Opus 5.5 at $4 per million input tokens and $20 per million output, and fast mode at $8 and $40.
Q5Where does inference run?
Anthropic lists US and Global endpoints. The location stays global, and US-only inference is set per request with inference_geo set to us.