GPT-5.6 Sol gets a new deployment path on Amazon Bedrock

On August 20, AWS said Amazon Bedrock began offering GPT-5.6 Sol, Terra, and Luna in more than 25 Regions through cross-Region inference. For Sol, the relevant identifiers are inference profiles such as `us.openai.gpt-5.6-sol` and `global.openai.gpt-5.6-sol`, rather than a raw model ID alone. AWS describes the feature as a capacity mechanism: a request submitted in one source Region can be processed by compute in an eligible destination Region, which broadens the available capacity pool and is intended to improve throughput and consistency under load.

The timing matters because OpenAI's June 26 GPT-5.6 Sol announcement described the family as a limited preview for selected partners, with broader availability planned later. AWS's August post is therefore a concrete cloud-distribution update for an already announced model family, not evidence of a separate model generation or a new benchmark result. It also puts the three GPT-5.6 tiers on a common Bedrock deployment surface, where AWS says they accept text and image input, return text, and support a one-million-token context window, reasoning mode, server-side tool calling, and prompt caching.

The practical distinction is between geographic and global routing. A geographic profile keeps processing within its predefined geography, while a global profile can route to supported commercial AWS Regions according to real-time capacity. That makes the global option useful when capacity and cost flexibility matter more than processing location, but it is not interchangeable with a residency-constrained deployment. AWS's general documentation says global profiles may process requests in any supported commercial Region, whereas geographic profiles are the option intended for compliance requirements tied to a geography.

For teams already using OpenAI-style clients, AWS says Bedrock exposes compatible Responses and Chat Completions APIs, while its own Converse API offers a Bedrock-native path. The migration is still not just an endpoint swap: applications must select an inference profile, configure IAM for the profile and its eligible model destinations, and account for routing in controls and observability. AWS says CloudTrail records the processing Region in `additionalEventData.inferenceRegion`; that is useful for audit trails, while the need to choose geographic routing remains the key limitation for workloads whose data cannot be processed outside a defined boundary.