Introducing OpenAI models on Amazon Bedrock for in-country inferencing in India

Amazon Bedrock now supports OpenAI’s GPT-5.6 Terra and Luna in India, and AWS is pushing hard on the part that will matter most to banks, hospitals, and public-sector teams: the requests stay inside the country. The setup uses India geographic cross-Region inference, so Bedrock can move traffic between Asia Pacific (Mumbai) and Asia Pacific (Hyderabad) without sending processing outside India.

The pitch here is capacity as much as compliance. Cross-Region inference spreads requests across a broader pool of compute instead of pinning everything to one Region, which should help when demand spikes. With the India profiles, that routing is limited to ap-south-1 and ap-south-2, and the inference profile itself is what you call from your application.

Both models support a 1-million-token context window, accept text and image input, and return text output. That makes them usable for long documents, large code bases, and mixed media requests in a single call. AWS also says billing, quota tracking, and logs stay anchored in the source Region, while CloudWatch and CloudTrail records remain in one place.

For developers, Bedrock exposes the models through the console and through code. AWS says new apps should use the bedrock-runtime endpoint, which supports the Bedrock-native InvokeModel and Converse APIs as well as the OpenAI-compatible Responses and Chat Completions APIs. The company also shows examples for server-side state, streaming, and prompt caching, including cached reads billed at a 90 percent discount versus uncached input tokens.

There’s also a broader global option if local residency is not the point. Bedrock’s global cross-Region inference can route GPT-5.6 traffic to commercial AWS Regions worldwide, including Sol, Terra, and Luna. But if the workload has to stay in-country, AWS is clear: use the India-prefixed profiles.