Claude Models on Google Cloud Guide

If your company already runs on Google Cloud, you can use Claude there too. You get Google Cloud billing, IAM permissions, quotas, and logging, while calling Claude with Anthropic’s own SDKs.

Same Claude, your existing cloud: one bill, one set of permissions.

Google managed

Runs on Google Cloud’s Agent Platform (formerly Vertex AI), billed by Google, through Model Garden

Anthropic managed

Run by Anthropic on Google Cloud, with a separate client and more Claude API features

This guide covers the Google managed version, the one most Google Cloud teams start with.

Claude Fable 5.1

claude-fable-5-1

Claude Opus 5.5

claude-opus-5-5

Claude Sonnet 5.5

claude-sonnet-5-5

Claude Haiku 5.5

claude-haiku-5-5

Claude Mythos 5.1

claude-mythos-5-1, limited availability

Older models are also listed. Availability differs by region, so search “Claude” in Model Garden to see what your region offers.

• A Google Cloud project with Agent Platform enabled
• The gcloud CLI installed for local work
• An IAM role with the aiplatform.endpoints.predict permission, such as Agent Platform User
• Credentials set up, for example with gcloud auth application-default login

Python

pip install -U "anthropic[vertex]"

TypeScript

npm install @anthropic-ai/vertex-sdk

SDKs also exist for C#, Go, Java, PHP, and Ruby.

In Python, create an AnthropicVertex client with your project ID and region set to “global”, then call messages.create with model “claude-opus-5-5”, a max_tokens value, and your messages. The rest works like the normal Messages API.

With plain HTTP, requests go to an Agent Platform URL ending in :rawPredict, or :streamRawPredict for streaming. The model goes in the URL, and anthropic_version goes in the request body.

Global

Recommended: highest availability, routes to any region with capacity

Multi-region

US or EU only, for broad data residency

Regional

One named region, for strict residency; Sonnet 4.6 and older models only

Multi-region and regional endpoints cost 10 percent more than global for recent models.

• Messages API, streaming, prompt caching, and extended thinking
• Tool use, including bash, computer use, browser use, memory, text editor, and web search tools
• Citations and structured outputs
• A 1M-token context window on recent models such as Opus 5.5, Sonnet 5.5, and Haiku 5.5

• The Files API and URL sources for images and documents
• Code execution, web fetch, and advisor server tools
• Agent Skills, the MCP connector, and programmatic tool calling
• Message Batches, Models, Admin, and Usage APIs
• Claude Managed Agents

Requests are capped at 30 MB, so very large files can hit that limit before the token limit.

• Google Cloud bills you; check Google’s generative AI pricing page for Claude prices
• Google manages rate limits; request more quota in the console
• Turn on request-response logging and keep at least 30 days of logs
• Google sets retirement dates, which can differ from Anthropic’s

Haiku 5.5

Fast, low-cost tasks such as classification and routing

Sonnet 5.5

Most everyday work and agents

Opus 5.5

Hard reasoning, coding, and long documents