Unlocking the Future of AI: Introducing Anthropic’s Claude Sonnet 4.5 and Haiku 4.5 on Amazon Bedrock in Japan and Australia
Enhancing Local Data Processing with Cross-Region Inference Capabilities
Unlocking Potential: Anthropic’s Claude Sonnet 4.5 and Claude Haiku 4.5 Now Live on Amazon Bedrock
こんにちは, G’day!
The world of generative AI has taken a significant leap forward with the recent launch of Anthropic’s Claude Sonnet 4.5 and Claude Haiku 4.5 on Amazon Bedrock. These cutting-edge models are engineered for complex agentic tasks, coding, and enterprise workloads, significantly empowering developers. In an exciting turn of events, customers in Japan and Australia can now utilize these models while ensuring data processing remains within geographical boundaries via Cross-Region inference (CRIS).
In this blog post, we will delve into the details of the geographic-specific CRIS profiles available in Japan and Australia, guide you through migrating from older models, and help you kickstart your journey to harness these advanced capabilities in your generative AI applications.
Japan and Australia Cross-Region Inference
The CRIS feature allows users in Japan and Australia to process requests locally. By employing the CRIS system, Amazon Bedrock ensures that inference requests are managed wholly within the designated geographic areas—either Japan or Australia.
How Cross-Region Inference Works
Operated through the AWS Global Network, CRIS ensures end-to-end encryption for data in transit and at rest. When an inference request is made, Amazon Bedrock assess available capacity within potential destination regions and directs the request to the optimal location. This routing takes place over the secure AWS Global Network, avoiding public internet traversals between regions listed as potential destinations.
In Japan, requests seamlessly flow between Tokyo and Osaka, while in Australia, routing occurs between Sydney and Melbourne. This automated intelligent routing ensures optimal traffic distribution without requiring manual intervention.
Cross-Region Inference Configuration
For organizations operating within Japan:
| Source Region | Destination Region | Description |
|---|---|---|
| ap-northeast-1 (Tokyo) | ap-northeast-1 (Tokyo), ap-northeast-3 (Osaka) | Requests can be routed to either Tokyo or Osaka Regions. |
| ap-northeast-3 (Osaka) | ap-northeast-1 (Tokyo), ap-northeast-3 (Osaka) | Requests can be routed to either Tokyo or Osaka Regions. |
For organizations in Australia:
| Source Region | Destination Region | Description |
|---|---|---|
| ap-southeast-2 (Sydney) | ap-southeast-2 (Sydney), ap-southeast-4 (Melbourne) | Requests can be routed to either Sydney or Melbourne Regions. |
| ap-southeast-4 (Melbourne) | ap-southeast-2 (Sydney), ap-southeast-4 (Melbourne) | Requests can be routed to either Sydney or Melbourne Regions. |
Getting Started
To initiate your experience with CRIS in Japan or Australia, follow these steps:
-
Configure IAM Permissions: Ensure that your IAM user or role has the necessary permissions to invoke Amazon Bedrock models. Specific policies will be required for geographic-specific inference profiles.
-
Use a Cross-Region Inference Profile: Configure your application to use the relevant inference profile ID, applicable for both the InvokeModel and Converse APIs.
Example Code for Using the Converse API (Python):
import boto3
# Initialize Bedrock Runtime client
bedrock_runtime = boto3.client(
service_name="bedrock-runtime",
region_name="ap-northeast-1" # Your originating Region
)
# Define the inference profile ID
inference_profile_id = "jp.anthropic.claude-sonnet-4-5-20250929-v1:0"
# Prepare the conversation
response = bedrock_runtime.converse(
modelId=inference_profile_id,
messages=[
{
"role": "user",
"content": [{"text": "What is Amazon Bedrock?"}]
}
],
inferenceConfig={
"maxTokens": 512,
"temperature": 0.7
}
)
# Print the response
print(f"Response: {response['output']['message']['content'][0]['text']}")Quota Management
Understanding quota management is crucial when using CRIS. Quotas for geographic-specific CRIS are managed at the source region level, meaning that quota increases requested from one region apply exclusively to requests originating from that region.
Requesting a Quota Increase
Organizations should utilize the AWS Service Quotas console for requesting quota increases in their respective source regions, ensuring they consider their usage patterns to make accurate requests.
Quota Management Best Practices
- Request Increases Proactively: Based on anticipated workloads, it’s important to request quota increases early through the AWS Service Quotas console.
- Monitor Utilization: Implement monitoring tools like CloudWatch for real-time tracking of usage to avoid hitting quota limits.
- Test at Scale: Load testing is essential for understanding quota expectations under realistic conditions.
Migrating from Claude 3.5 to Claude 4.5
Organizations currently using Claude Sonnet 3.5 and Claude Haiku 3.5 should strategize their migration to the newer 4.5 models. The enhancements in Claude 4.5, particularly in memory management and task execution, make this migration a valuable opportunity.
Key Migration Considerations
- Performance Benchmarking: Establish benchmarks to evaluate whether the new model meets or exceeds current performance demands.
- Technical Considerations: Adjust existing prompts to utilize advanced context processing and tool handling.
- Cost Optimization: Assess potential costs associated with enhanced capabilities and quota management.
Choosing Between Global or GEO Cross-Region Inference
Customers can choose between Global CRIS—ideal for those without specific geographical data residency requirements—and GEO CRIS, which is recommended for processing data within stipulated geographic boundaries.
Conclusion
In this exploration of Anthropic’s Claude Sonnet 4.5 and Claude Haiku 4.5 on Amazon Bedrock, we’ve highlighted the enhanced capabilities available to customers in Japan and Australia through Cross-Region inference. This feature not only maximizes AI potential but also enables compliance with local data processing requirements, making it ideal for industries like finance and healthcare.
By following the steps provided and understanding the nuances of quota management and migration strategies, organizations can confidently implement these advanced AI models in their applications.
For detailed insights into pricing and further information on the models, visit the Amazon Bedrock pricing page.
Through these advancements, businesses can responsibly harness the power of AI technology while ensuring compliance with necessary regulations, paving the way for innovative and efficient applications.
About the Authors
- Derrick Choo – Senior Solutions Architect at AWS
- Melanie Li, PhD – Senior Generative AI Specialist Solutions Architect at AWS
- Saurabh Trikande – Senior Product Manager for Amazon Bedrock
- Jared Dean – Principal AI/ML Solutions Architect at AWS
- Stephanie Zhao – Generative AI GTM & Capacity Lead for AWS in Asia Pacific and Japan
- Kazuki Motohashi, Ph.D. – AI/ML GTM Specialist Solutions Architect at AWS Japan
Enjoy exploring the transformative capabilities of Anthropic’s models in Amazon Bedrock!