AWS Brings OpenAI's GPT-5.6 Models to Bedrock in India

Amazon Bedrock customers in India can now run OpenAI's GPT-5.6 models, Terra and Luna, without their data leaving the country. AWS has enabled the two models through dedicated inference profiles, identified as in.openai.gpt-5.6-terra and in.openai.gpt-5.6-luna, built specifically for the India geography.

The launch centers on a feature AWS calls India Geo cross-Region inference. Rather than pinning workloads to a single data center, it automatically distributes requests between AWS's Asia Pacific Mumbai and Asia Pacific Hyderabad Regions.

That routing exists for a practical reason: many Indian enterprises, particularly in regulated sectors like finance and healthcare, are bound by rules requiring that data processing stay within national borders. By confining inference to the two India Regions, Bedrock lets teams tap GPT-5.6 while keeping their compliance posture intact.

Developers get flexibility on the API side as well. Terra and Luna support Bedrock's Responses, Chat Completions, and Converse APIs, meaning teams already building against one of those interfaces can adopt the new models without rewriting their integration layer.

What operators can monitor

AWS paired the model availability with its standard operational tooling. Account-level controls now cover model invocation logging, which can be routed to either Amazon S3 or CloudWatch Logs, alongside CloudWatch metrics and itemized cost tracking.

For engineering teams, that combination matters as much as the model access itself. Logging and metrics give platform teams the audit trail needed to satisfy internal governance reviews, while cost itemization helps finance teams attribute GPT-5.6 usage to specific projects or business units.

The move fits a broader pattern of Bedrock expanding regional access to third-party frontier models rather than restricting them to AWS's home Regions in the US. For companies operating in India that have wanted to use OpenAI's models but faced data residency constraints, this removes a concrete blocker.

No pricing details specific to the India deployment were disclosed alongside the announcement. What is confirmed is the mechanism: two named models, two Regions handling the compute, and a logging and cost stack built to make the arrangement auditable from day one.