Region Settings for Using Claude on Vertex AI | global vs us & eu

What Are Vertex AI Claude Regions?

In Vertex AI, you specify a region to determine which geographically located servers receive your requests when calling Claude models. The region you choose affects the following:

• Latency: Regions closer to your infrastructure respond faster
• Data storage location: Critical for meeting industry regulations and compliance requirements
• Pricing: Regional endpoints carry a 10% surcharge compared to the global endpoint
• Throughput provisioning: Provisioned Throughput is only available on regional endpoints

Official references include Claude on Vertex AI — Claude Docs and the Google Cloud Blog.

Available Regions

Region options for calling Claude on Vertex AI fall into three broad categories.

Global Endpoint (global)

The simplest option. Set region="global" and Google dynamically routes your requests across multiple regions. Since data storage is not restricted to a specific geography, this is ideal for projects without strict data residency requirements. No pricing surcharge.

Multi-Region Endpoints (us / eu)

A new endpoint format announced by Google Cloud in 2025. These retain the flexibility of dynamic routing while keeping data within a specific geographic area.

Pricing is 10% higher than global. Useful for GDPR compliance and other EU data protection requirements, or when data must remain within the United States.

Single-Region Endpoints

Use these when you need to restrict requests to a specific region. Examples of available regions:

Single-region endpoints also carry a 10% pricing surcharge, the same as multi-region. Additionally, Provisioned Throughput is only available on single-region endpoints.

The Tokyo region (asia-northeast1) may not offer Claude models, so Singapore (asia-southeast1) is currently the recommended choice for Asia-Pacific deployments.

How to Configure Regions

Python (AnthropicVertex SDK)

from anthropic import AnthropicVertex

# Global endpoint (default)
client = AnthropicVertex(project_id="MY_PROJECT_ID", region="global")

# Multi-region US
client = AnthropicVertex(project_id="MY_PROJECT_ID", region="us")

# Multi-region EU
client = AnthropicVertex(project_id="MY_PROJECT_ID", region="eu")

# Single region (e.g., Eastern US)
client = AnthropicVertex(project_id="MY_PROJECT_ID", region="us-east5")

message = client.messages.create(
    model="claude-opus-4-8@20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
)

TypeScript (@anthropic-ai/vertex-sdk)

import { AnthropicVertex } from "@anthropic-ai/vertex-sdk";

// Multi-region EU (for GDPR compliance, etc.)
const client = new AnthropicVertex({
  projectId: "MY_PROJECT_ID",
  region: "eu",
});

const message = await client.messages.create({
  model: "claude-opus-4-8@20250514",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello" }],
});

cURL

MODEL_ID=claude-opus-4-8@20250514
LOCATION=us   # global / us / eu / us-east5, etc.
PROJECT_ID=MY_PROJECT_ID

# For multi-region US, the host differs from the standard endpoint
curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  https://aiplatform.us.rep.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/publishers/anthropic/models/${MODEL_ID}:streamRawPredict \
  -d '{
    "anthropic_version": "vertex-2023-10-16",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

For global or standard single-region endpoints, use {LOCATION}-aiplatform.googleapis.com.

Global vs. Multi-Region vs. Single-Region Comparison

For most development and production environments, global offers the best cost efficiency with the simplest configuration. A natural decision flow is to start with global and switch to us, eu, or a specific region only when compliance requirements arise.

Data Residency and Compliance

If you need GDPR compliance for European services, or if US financial or healthcare regulations require data to remain within the country, you will need to choose a multi-region or single-region endpoint.

• GDPR compliance: Specifying eu (or an EU region such as europe-west1) ensures that request and response data is processed only on servers within the EU.
• US-only data: Using us restricts routing to Google's US regions (such as us-central1 or us-east4).
• VPC Service Controls: Vertex AI can be combined with VPC Service Controls to allow requests only from within a specific network perimeter. This configuration requires single-region endpoints.

The global endpoint offers no guarantee of which country's servers process your data. If you have data residency requirements, always use a geographically scoped endpoint.

Pricing Impact

According to the Claude on Vertex AI official documentation, using multi-region or single-region endpoints results in a 10% surcharge over the standard price for the global endpoint.

For example, Claude Sonnet 4.5 on us or eu endpoints costs 1.1x the per-token rate compared to the global endpoint.

Note that some models released before the Claude 4 series follow a different pricing structure with a different surcharge rate, so check the pricing page for the specific model you are using.

If your primary goal is minimizing cost for traffic originating from Japan, global is the best choice. Selecting the APAC region (asia-southeast1) adds a 10% surcharge with limited latency improvement, so the decision between global and regional should be driven by whether you have data residency requirements.

Region and Model Availability

Not all models are available in every region. You can check model availability per region for your project in Google Cloud's Model Garden (console.cloud.google.com/vertex-ai/model-garden).

Shortly after a new model is added (such as Claude Opus 4.8), it may be available on global or us-east5 but not yet in other regions. When testing the latest model via Vertex AI, it is safest to start with global or us-east5.

Summary

When choosing a region for Claude on Vertex AI, the key decision factor is whether you have data storage requirements.

• No requirements → global for simplicity and lower cost
• EU data protection (GDPR, etc.) → eu multi-region
• US-only data → us multi-region
• Strict single-region control or Provisioned Throughput → single-region such as us-east5

Configuration is as simple as changing the region= parameter in one place, for both Python and TypeScript. Note that multi-region endpoints use different hostnames (aiplatform.us.rep.googleapis.com / aiplatform.eu.rep.googleapis.com). Choose the right endpoint by balancing compliance requirements against cost.