Amazon adds OpenAI GPT-5.6 models to its US government cloud
Requests go across two AWS GovCloud (US) Regions and stay in the US. Amazon gave no speed numbers.
Amazon Bedrock, an Amazon service that gives access to AI models, holds the OpenAI GPT-5.6 models Terra and Luna in AWS GovCloud (US). This is the Amazon cloud for US government work. It has three APIs (software links to the models): Responses, Converse and Chat Completions. It also adds cross-Region inference, which sends each request to the Region that has capacity.
What is new
Cross-Region inference sends requests across the two GovCloud (US) Regions. Amazon tells that this gives higher throughput and more redundancy, and it keeps data in the US. The customer does not manage capacity in each Region.
The controls
The OpenAI models use the same account controls as other Bedrock models. Amazon lists these:
- Invocation logs, sent to Amazon S3 or CloudWatch Logs.
- CloudWatch metrics: calls, tokens, latency, throttles and errors.
- Cost reports that show spend for each model.
What is not known
Amazon gave no test results, no price and no figure for the increase in throughput. The text does not say which government groups use the models.
Insight: OpenAI reaches government buyers through Amazon
The fast Decisions API of OpenAI uses Luna. OpenAI opened it to all customers on 22 October 2026. Luna and Terra also operate in the Amazon government cloud, with Amazon logs and bills. This reads as OpenAI reaching government buyers through an Amazon gate, where Amazon keeps the control and cost data. Monitor if the Decisions API follows to GovCloud.
Sources
Posted