How to install Introducing India cross-Region inference for OpenAI

What happened

Amazon Bedrock has expanded its model lineup to include the OpenAI GPT-5.6 models, Terra and Luna, now available in India through India geographic cross-Region inference. This addition lets teams working under Indian data residency requirements access these models at scale without routing inference traffic outside the country.

If your organization processes sensitive data subject to local regulations, this release removes a previous obstacle to adopting these models in production. Amazon Bedrock handles the infrastructure to keep inference requests and data within India, so you can focus on building rather than on compliance workarounds.

Amazon Bedrock now offers two OpenAI GPT-5.6 models under India geographic cross-Region inference: Terra and Luna. Both models are accessible through the standard Bedrock API surface. Inference requests made from supported India regions stay within India at the infrastructure level, with Bedrock managing regional routing automatically on your behalf.

How it works

The primary change is geographic availability, not the API itself. If you were already calling Terra or Luna through Bedrock in another geography, the request shape, authentication method, and response format remain the same. The difference is that you now specify an India region as your endpoint and Bedrock ensures the request does not leave Indian infrastructure. Teams that previously could not use these models because of data processing requirements now have a supported path to do so. No new SDK version is required to access the feature, but you must target an India region endpoint when constructing your client.

First, confirm you have the AWS CLI or the relevant AWS SDK installed and configured with credentials that have Bedrock permissions in your target India region. If you are starting fresh, install the AWS CLI by following the official AWS documentation for your operating system, then run the configuration step to set your access key, secret, and default region to the India region where you want to work.

If you are upgrading an existing integration, update your SDK package to the latest published version using your package manager. For Python this means running pip install with the upgrade flag against boto3. For Node.js, run npm install with the latest flag against the aws-sdk or the modular client package you use. For Java or other languages, update the dependency version in your build file and re-resolve dependencies.

Why it matters

After updating, change the region value in your Bedrock client constructor or your environment configuration to the India region. When you make an InvokeModel or Converse API call, set the model identifier to either the Terra or Luna model ID as listed in the Bedrock console under the India region. Submit a test request with a short prompt to confirm routing is working and that you receive a response. Check the response metadata or your AWS CloudTrail logs to verify the request was processed in the expected region.

If you use infrastructure-as-code tools such as AWS CloudFormation or Terraform, update the region parameter in your Bedrock resource definitions and redeploy. No special resource type is needed beyond the standard Bedrock model invocation permissions.

Cross-Region inference in this context means Bedrock routes requests within India across its India infrastructure, not that it routes between India and other geographies. Do not confuse this with a feature that blends Indian and non-Indian compute. All data stays within India.

Who is affected

Your IAM policies must grant Bedrock invocation permissions scoped to the India region. A policy that grants access only in us-east-1 or another non-India region will not work here. Review your resource-based and identity-based policies before testing.

Model availability in the Bedrock console is region-specific. If you log in and set your console region to something other than the India region, you will not see Terra and Luna listed under the India cross-Region inference feature. Switch your console region before checking availability.

Existing applications that call Bedrock in other regions are not automatically migrated. You must explicitly reconfigure endpoints and region settings. Running two configurations in parallel, one for an existing region and one for India, is straightforward as long as each client instance targets the correct region.

What to watch next

Throughput limits and quota allocations are region-specific. If you hit throttling after switching regions, request a quota increase in the India region through the AWS Service Quotas console rather than in your previous region.

Monitor the Bedrock model catalog in the India region for any additions to the cross-Region inference lineup, as AWS tends to expand regional availability of models incrementally. Keep an eye on the AWS Machine Learning Blog and the Bedrock service release notes for updates to quotas, pricing for India region inference, and any new compliance certifications that may further simplify regulatory approval for your use case.