GPT-5.6 Sol gains cross-Region routing on Amazon Bedrock

AWS says that Amazon Bedrock now offers GPT-5.6 Sol, alongside Terra and Luna, in more than 25 AWS Regions through cross-Region inference. The mechanism is an inference profile: an application calls the profile from a source Region, and Bedrock can route the request to eligible destination-region compute. AWS describes this primarily as a capacity mechanism, intended to draw on a broader compute pool when a single Region is constrained. For GPT-5.6 Sol users, the practical change is a new Bedrock deployment option rather than a newly documented model behavior or benchmark result.

The announcement was published by AWS on August 20 and was co-written with an OpenAI contributor. AWS identifies Sol, Terra, and Luna as the three general-purpose GPT-5.6 variants covered by the cross-Region rollout. It says those variants accept text and image inputs, return text, and support a one-million-token context window, reasoning mode, server-side tool calling, prompt caching, and streaming. They can be invoked through the OpenAI Responses API, the OpenAI Chat Completions API, or Bedrock's Converse interface, so the routing change can fit either OpenAI-compatible or Bedrock-native integrations.

Compared with a direct single-Region invocation, a cross-Region profile makes routing policy explicit in the model identifier. A geographic profile such as the US option confines processing to the profile's stated geography, while a global profile can use supported commercial AWS Regions worldwide according to capacity. AWS documentation frames the tradeoff as compliance and residency control versus wider routing and lower global-profile pricing, and it notes that inference profiles do not support Provisioned Throughput. That means the feature is not a substitute for every capacity-planning path, even though it can increase the available on-demand pool.

Technical teams should treat profile selection as an architecture and governance decision. AWS says global routing may process data in any eligible Region, while geographic profiles retain it within the selected boundary; CloudTrail records the processing Region in additional event data. The documentation also says cross-Region traffic stays on the AWS network and is encrypted in transit, but AWS notes that content flagged by Bedrock abuse-detection classifiers for GPT-5.6 can be retained for up to 30 days for offline abuse detection. Teams with residency requirements, restrictive service-control policies, or strict retention reviews should validate the profile's routing set and IAM permissions before switching production traffic.