OutYet reporting
AWS adds a cross-Region path for GPT-5.6 Sol, but the endpoint details need care
AWS says GPT-5.6 Sol can use geographic and global Bedrock inference profiles, while its Sol model card still documents a separate in-Region endpoint.
Amazon Web Services says OpenAI GPT-5.6 Sol, Terra, and Luna can now be called in Amazon Bedrock through US geographic and global cross-Region inference profiles. AWS describes an inference profile as a capacity mechanism: an application submits work from a source Region and Bedrock can route it to an eligible destination Region rather than confining it to one Region. The provider says the three models accept text and image inputs, return text, and support the Responses, Chat Completions, and Converse APIs. This is a deployment and routing change for the GPT-5.6 family, not evidence of a new model.
The timeline is short. AWS posted a product update on August 17 saying Bedrock had added GPT-5.6 support on bedrock-runtime, along with Geo and Global cross-Region inference. Its August 20 technical post then named the profile conventions, including `us.openai.gpt-5.6-sol` and `global.openai.gpt-5.6-sol`. AWS says the US geographic option keeps processing within its defined geography, whereas the global option can route work among supported commercial Regions. Those are provider statements about service behavior, not independent measurements of throughput or latency.
The practical complication is that AWS's GPT-5.6 Sol model card documents a different path. That card directs Sol requests to the bedrock-mantle endpoint, lists the raw `openai.gpt-5.6-sol` model ID, and marks Geo and Global inference IDs as unsupported, with in-Region availability shown for us-east-1 and us-east-2. The newer AWS announcement instead describes bedrock-runtime and cross-Region profiles. Taken together, the pages suggest distinct access paths or a documentation transition, but the Sol card does not explain the difference. Teams should validate the endpoint, model or profile ID, and Region configuration in their own account before changing production routing.
For users already operating on Bedrock, the change could provide a second capacity pool and an OpenAI-compatible integration route without moving workloads to a separate provider account. It does not remove governance work. AWS says model access must be enabled, geographic profiles require permissions across their destination Regions, and global profiles interact with Region-restrictive service control policies differently. AWS also says geographic and global profiles have separate quota allocations, so routing choice affects the quota pool a workload uses. The useful test is therefore operational, not promotional: confirm the allowed data geography, IAM and SCP behavior, quota assignment, and observed cost and latency for the exact profile.