What AWS announced
AWS’s October 30 announcement names GPT-5.6 Terra and Luna for Responses, Chat Completions, and Converse on bedrock-runtime in AWS GovCloud (US). It says cross-region inference routes across both GovCloud regions and connects runtime requests to logging, metrics, and spend attribution. [1]
Current model cards confirm the API support generally, but their examined GovCloud tables still show bedrock-mantle, without GovCloud runtime profile details. Check the current guide and applicable permissions before changing an integration. The notice does not establish access for a particular account. [2][3]
Our take
Treat an endpoint change as a routing and governance change. Match data boundaries, permissions, and observability to your requirements, then validate the workflow before moving production traffic.