AWS Brings GPT-5.6 Cross-Region Inference, With a Data-Location Choice

Amazon Bedrock now offers OpenAI’s GPT-5.6 Sol, Terra, and Luna through cross-Region inference in more than 25 AWS Regions. Developers can select a US profile that keeps processing within that geography or a global profile that routes to available capacity across supported commercial Regions. The choice turns data-location requirements into an operational constraint on capacity.

The feature works through inference profiles: logical identifiers passed in place of a raw model ID. An application sends its request from a source Region, and Bedrock can process it in a permitted destination Region. AWS describes the system as a capacity mechanism, designed to draw on a broader compute pool and help maintain throughput under load rather than leave a workload bound to one Region’s capacity.

That distinction matters most for workloads with residency rules. AWS says customers whose processing must remain in a specified geography should use the corresponding geographic profile or call a single Region directly; the global option is for workloads without geographic processing requirements. Billing and quota consumption remain associated with the customer account regardless of the backend Region that handled a request.

A model family built for existing application paths

The three general-purpose variants accept text and image inputs, return text, and list a one-million-token context window. They support reasoning mode, server-side tool calling, and prompt caching. AWS says Sol, Terra, and Luna are tuned for different capability-and-cost balances, but this launch’s routing rules apply to all three.

Integration options

• Use the OpenAI Responses API or Chat Completions API through Bedrock’s OpenAI-compatible endpoint, replacing the model parameter with a geographic or global inference-profile ID.
• Use Bedrock’s Converse API or ConverseStream when an application already uses Bedrock’s common interface for model calls and streaming.
• Authenticate with AWS credentials or a Bedrock API key; AWS recommends programmatically generated short-term keys for production rather than long-term keys intended for exploration.

The routing choice reaches security operations

Cross-Region access requires IAM permissions for both the inference profile and the foundation model in every Region that profile may use. Requests appear in CloudTrail in the source Region, and an inferenceRegion field identifies the Region that processed each request. AWS also says calls can reach Bedrock privately through a VPC endpoint and remain governed by IAM policies.

One retention condition deserves separate review. For GPT-5.6, AWS says content flagged by its automated abuse-detection classifiers may be retained for up to 30 days for offline abuse detection. That condition sits alongside AWS’s stated zero-operator-access security model, which it says prevents AWS operators from accessing prompts or completions.

For teams already using OpenAI request formats, the practical change can be narrow: point the client at Bedrock and choose the profile ID that matches the workload’s processing boundary. The harder decision is not API syntax, but whether a wider capacity pool is compatible with the application’s data-handling rules.