OpenAI GPT-5.6 Terra and Luna reach Bedrock in AWS GovCloud with cross-region inference

OpenAI models are now available to US government customers through AWS. Bedrock serves GPT-5.6 Terra and Luna on the bedrock-runtime endpoint in AWS GovCloud (US), with support for OpenAI's Responses API as well as Bedrock's Converse and Chat Completions APIs. Invocations work with Bedrock model logging, CloudWatch metrics and Cost Explorer.

The launch adds GovCloud Cross-Region Inference, which dynamically routes requests across GovCloud regions for higher throughput and redundancy without leaving US sovereign boundaries. That addresses the rate limits and capacity spikes that typically hurt public-sector deployments.

AWS paired this with a guide for routing the ChatGPT desktop app's inference to Bedrock in GovCloud (US-West). The setup uses API-mode sign-in, a managed device policy, a named AWS profile and network restrictions, so security teams can validate the app before a wider rollout. That means agencies can give staff a familiar ChatGPT front end while inference stays inside their AWS boundary.

The move continues a striking shift: OpenAI models now flow through AWS, long Microsoft's exclusive partner territory. Matt Wood announced GPT-6.1 Terra on commercial Bedrock on October 26. The GovCloud versions trail the commercial GPT-6 family, which is typical given accreditation timelines. Watch for FedRAMP High and IL-level authorizations for newer models.