AWS has enabled access to Anthropic Claude Opus 5, Sonnet 5, and Haiku 4.5 models via the Amazon Bedrock India regional endpoint. The service uses geographic cross-region inference, routing requests between the Mumbai and Hyderabad regions to manage capacity while ensuring data remains within Indian borders to meet local processing requirements.
Local data routing and capacity management
AWS introduced a geographic cross-region inference profile specifically for the India market. This system automatically distributes model requests between the Mumbai (ap-south-1) and Hyderabad (ap-south-2) regions. The setup is designed to draw from a broader pool of compute resources than a single region could provide, aimed at maintaining consistent throughput during high traffic periods.
The infrastructure operates through the bedrock-runtime endpoint and supports the Anthropic Messages API alongside native Amazon Bedrock APIs. By keeping inference traffic within India, the service addresses specific organizational needs for processing data within a defined geography. All data in transit is encrypted, and billing and logging remain centralized in the source region for monitoring purposes.
Security protocols and availability limits
Under the Amazon Bedrock zero data retention model, inputs and outputs are not stored in destination regions. However, the service includes a condition where human review by AWS may occur if automated safety classifiers flag specific content. Users can currently access the Claude Opus 5, Sonnet 5, and Haiku 4.5 variants through the Amazon Bedrock console playground or via SDKs.
While the system provides broader capacity, it is limited to routing between two specific Indian regions. The release focuses on inference rather than model training or fine-tuning. The effectiveness of the automated routing in preventing latency during simultaneous peak loads across both Indian regions remains to be observed in large-scale production environments.
Original source
This report summarises the source below. Analysis is labelled separately; product and research claims remain attributed to their source.
Read the original at AWS ML