AWS detailed global cross-Region inference for GPT-6 Sol, Luna and Astra on Bedrock, including supported source Regions and operational requirements.
Amazon Web Services published guidance on October 12 for invoking OpenAI’s GPT-6 models through global cross-Region inference on Amazon Bedrock. GPT-6 Sol and GPT-6 Luna can be called from AWS Regions in Singapore, Jakarta, Malaysia, Thailand and Taipei; GPT-6 Astra is available through this setup from Singapore. AWS says requests are processed in a supported commercial Region where the model is deployed, selected by Bedrock.
The post describes three model options: Sol for reasoning and coding, Luna for focused tasks at high volume, and Astra for complex reasoning, coding and long documents. All three accept text and image inputs, client-side tool use and prompt caching. Developers can call them using Bedrock’s Converse API or OpenAI-compatible Responses and Chat Completions APIs, passing a global inference profile ID as the model identifier.
Cross-Region routing changes where processing occurs, but AWS says billing, quota tracking, CloudTrail events and CloudWatch metrics remain associated with the source Region. CloudTrail can record the processing Region for an individual request. The post also cautions that destination Regions are not limited to the caller’s country, a consideration for organizations with data-residency requirements.
AWS’s setup guidance covers IAM permissions and service control policies, including how Region allowlists can block requests unless an exception is scoped to the Bedrock inference profile. It also says classifier-flagged traffic may be retained for automated abuse detection for up to 30 days in the destination Region; eligible customers may ask their AWS account team about full zero data retention.
Quota planning is another operational detail: GPT-6 quotas are measured in tokens per minute against the source Region, and each output token consumes 10 tokens of quota. AWS advises teams to plan capacity and monitor usage before production deployment. The post presents this as a configuration path for teams seeking access to Bedrock capacity beyond a single Region’s limits.