Bedrock's OpenAI Models Get a Throughput and Cost Boost with Cross-Region Inference!

Hey there, it's me! Today I found a nice update around the OpenAI models on Amazon Bedrock, so let me share it with you.

AWS What's New aws.amazon.com

What was announced?

Over on AWS What's New, AWS announced expanded API support for OpenAI models on Amazon Bedrock, plus the introduction of Cross-Region Inference.

This covers OpenAI's GPT-5.6 model family (Sol, Terra, and Luna). On the bedrock-runtime endpoint, they now support three API formats:

• Responses API
• Converse API
• Chat Completions API

On top of that, because these native OpenAI APIs now run on bedrock-runtime, you get the same account-level controls used for other Bedrock models, including Bedrock model invocation logging.

The story so far

Until now, OpenAI models weren't available as native APIs on the bedrock-runtime endpoint, and there was no support for routing inference requests across regions. Requests were basically handled within a single, fixed region.

What changes

The headline feature here is Cross-Region Inference, and you now get to pick between two routing styles depending on what you need.

• Global cross region inference: routes requests across any commercial AWS Region where the model is available. This helps you keep throughput high during demand spikes, and it's also priced lower per token than in-Region or Geo inference
• Geo cross region inference: routes requests only within a predefined geography. This update adds new support for US Geo (US CRIS), which is a good fit if you need to keep data within a specific geography

In short, pick Global when throughput and cost matter most, and Geo when you need to keep the data's location constrained.

Since this now runs through Bedrock's own plumbing, tracking usage and logs for these models works the same way it does for other Bedrock models, which is a nice bonus.

Dive Deep

To get started, AWS recommends checking the model cards for GPT-5.6 in the documentation, which list the supported APIs, regions, and pricing details for each model.

On pricing specifically:

• Global cross region inference is priced lower per token than in-Region inference or Geo cross region inference
• Geo cross region inference routes within a geography, so it's more about data residency than squeezing out the lowest cost

Wrap-up

• OpenAI's GPT-5.6 models (Sol, Terra, and Luna) on Amazon Bedrock now support the Responses, Converse, and Chat Completions APIs
• You can now use Cross-Region Inference, choosing between Global (broad routing, high throughput, lower cost) and Geo (geography-scoped, now including US Geo)
• Native bedrock-runtime support means you get the same account-level controls and logging as other Bedrock models
• Check the GPT-5.6 model cards for the full details

This update really lands for engineers running OpenAI models on Bedrock in production who care about throughput, cost, and where their data lives!