AWS lists GPT-6 Sol and GPT-6 Luna for Amazon Bedrock global cross-Region inference from five Asia-Pacific source Regions, and GPT-6 Astra from Singapore. In guidance dated October 12, 2026, AWS says Bedrock selects a supported destination Region for processing, so the Region that sends a request does not determine where inference runs. AWS’s regional access and configuration guidance describes the profiles and their routing.
The GPT-6 profiles AWS lists
| Model | Global inference profile ID | AWS-listed source Regions |
| GPT-6 Sol | global.openai.gpt-6-sol | Singapore (ap-southeast-1), Jakarta (ap-southeast-3), Malaysia (ap-southeast-5), Thailand (ap-southeast-7), Taipei (ap-east-2) |
| GPT-6 Luna | global.openai.gpt-6-luna | Singapore (ap-southeast-1), Jakarta (ap-southeast-3), Malaysia (ap-southeast-5), Thailand (ap-southeast-7), Taipei (ap-east-2) |
| GPT-6 Astra | global.openai.gpt-6-astra | Singapore (ap-southeast-1) |
These are AWS Regions from which callers can invoke the profiles through Bedrock. Singapore is enabled by default in AWS accounts; Jakarta, Malaysia, Thailand and Taipei must be activated as source Regions first. AWS says destination Regions do not need to be activated by the caller.
Where Bedrock processes a request
A source Region is where the caller makes the API request. Bedrock routes it to a supported commercial AWS Region where the model is deployed. That means choosing a source Region in a particular country does not ensure that inference or related data handling stays there.
AWS keeps billing, quota tracking, CloudWatch metrics and CloudTrail events associated with the source Region. Classifier-flagged traffic used for automated offline abuse detection may be retained for up to 30 days in the destination Region. Eligible customers may request full zero data retention through their AWS account team.
For successful requests, AWS identifies additionalEventData.inferenceRegion in CloudTrail as the processing Region. CloudTrail Event History retains management events for 90 days. That record can help teams check where a particular request was processed.
Calling the profiles
AWS lists Bedrock Converse, OpenAI Responses and OpenAI Chat Completions as invocation options. The OpenAI-compatible APIs use /openai/v1; requests pass the relevant profile ID as the model and can authenticate with Signature Version 4 (SigV4) or an Amazon Bedrock API key.
This regional inference feature is separate from Bedrock Managed Agents, which NeoTeo covered in an earlier explainer.
