AWS adds cross-Region capacity to GPT-5.6 Sol, with governance tradeoffs

Amazon Bedrock has added cross-Region inference for OpenAI's GPT-5.6 family, including GPT-5.6 Sol. AWS says the new profiles can be invoked from more than 25 AWS Regions and offer both US-geographic and global routing. This is a serving and capacity change, not an assertion about new model weights: Bedrock continues to expose the Sol, Terra, and Luna variants while an inference profile lets AWS route a request to an allowed destination Region.

That makes the August change materially different from the earlier Bedrock footprint. AWS's July service update documented Sol in US East (N. Virginia) and US East (Ohio), while Terra and Luna were also documented in US West (Oregon). The new cross-Region configuration keeps the same family but supplies a 1 million-token context window, reasoning mode, server-side tool calling, prompt caching, and OpenAI-compatible Responses and Chat Completions APIs alongside Bedrock Converse. It therefore broadens deployment options more than it changes the programming model.

Profile selection is the actual architecture decision. AWS states that a US geographic profile keeps processing within its defined geography, whereas a global profile may use any supported commercial Region based on capacity. The documentation warns that prompts and outputs can move outside the caller's source Region, and AWS says content flagged by automated abuse detection for GPT-5.6 can be retained for as long as 30 days. Teams subject to residency or retention obligations should treat the global profile as an explicit data-governance choice, not merely an availability setting.

Migration can be small in code but larger in cloud policy. AWS shows an OpenAI SDK client pointed at Bedrock's /openai/v1 endpoint with a geographic or global inference-profile ID in place of the raw model ID and permits either AWS credentials or Bedrock API-key authentication. Yet a role must be allowed to invoke the profile and the foundation model in every destination Region, and region-restrictive service control policies can make cross-Region calls fail. AWS records the processing Region in CloudTrail, so operators should test the selected profile, inspect inferenceRegion, and align IAM and SCP rules before relying on it for production burst capacity.