Amazon Bedrock expands Claude model availability to in-country inferencing in India

AWS has turned on Anthropic’s Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5 for customers in India, and the pitch is simple: keep inference in-country while still leaning on cross-Region capacity. The new India regional endpoint uses geographic cross-Region inference, so requests from a source Region can be routed between ap-south-1 and ap-south-2 instead of spilling out to a wider set of locations.

That matters for anyone trying to satisfy local data-processing requirements without giving up scale. AWS says the setup lets requests draw from a broader compute pool, which should help with throughput and steadier performance when traffic spikes. The company also says the secure AWS network handles the traffic with end-to-end encryption in transit.

The data story is more specific than the usual cloud marketing fog. AWS says customer data is not stored in a destination Region when cross-Region inference is used, and that Amazon Bedrock uses a zero data retention model by default, meaning it does not store model inputs or outputs unless a flagged case requires human review as part of AWS safety checks.

Billing and quota usage stay tied to the source Region, and logs in CloudWatch and CloudTrail are recorded there too. The feature is available on the bedrock-runtime endpoint, and it works with Anthropic’s Messages API as well as Amazon Bedrock’s InvokeModel and Converse APIs. AWS also says it supports Bedrock Guardrails and intelligent prompt routing.

For people who just want to poke at the models first, the Bedrock console route is there too: open the console in the Region you want as the source, go to Playground, choose a Claude model, apply the India inference profile, and run a prompt. The post also walks through code examples using Boto3, the Anthropic SDK and the Bedrock token generator, which is AWS doing what AWS always does — turning a policy promise into three SDKs and a long afternoon.