Introducing cross-Region inference for OpenAI GPT-5.6 models on Amazon Bedrock
Amazon Web Services Melanie Li
AWS added OpenAI GPT-5.6 to Bedrock across 25+ Regions with cross-Region routing. It’s meant to smooth out capacity and keep data inside a geography when needed.
Based on reporting by Amazon Web Services, Melanie Li — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Amazon Bedrock now has OpenAI GPT-5.6 models in more than 25 AWS Regions, and the hook is cross-Region inference. That means a request no longer has to sit and wait on one Region’s spare capacity; Bedrock can send it somewhere else in the same geography, or, for the global profile, across supported commercial Regions where the model is deployed.
The rollout covers three general-purpose GPT-5.6 variants: Sol, Terra and Luna. They all take text and image inputs, return text, and come with a 1 million token context window. They also support reasoning mode, server-side tool calling and prompt caching. If you want to try them without wiring up code first, AWS points you to the Bedrock console’s playground, where the model picker shows both the geographic and global profiles.
Cross-Region inference here is driven by inference profiles, which are logical identifiers rather than raw model IDs. A geographic profile keeps processing inside a specific boundary, such as the US set, which matters if data residency is part of the job. A global profile can route more broadly, based on real-time capacity, and AWS says billing and quota are still tracked to your account no matter which backend Region actually handles the call.
The API story is unusually broad. GPT-5.6 on Bedrock speaks the OpenAI Responses API natively, and the same client can also use Chat Completions. If you’d rather stay inside Bedrock’s own interface, Converse works too, including streaming. AWS even shows the models being called from the OpenAI SDK pointed at Bedrock’s compatible endpoint, which is a neat little reminder that the integration is supposed to feel boring, not heroic.
The security and compliance pitch stays familiar: IAM controls access, calls can go through a VPC endpoint, CloudTrail logs them, and Bedrock’s zero-operator model is still in place. For some models, including GPT-5.6, content flagged by abuse-detection classifiers can be retained for up to 30 days for offline abuse detection. The practical message is simple: AWS is trying to make capacity elasticity look like an ordinary model setting, not a special deployment project.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI infrastructure news: less glitter, more plumbing. OpenAI on Bedrock with routing profiles is basically AWS saying capacity and residency should be configuration, not drama. The funny part is that the real innovation here is making the whole thing sound boring, which is usually how enterprise buyers know it might actually work.
Read more about this at: Amazon Web Services