The Pulse
AWS Adds In-Region Claude 5 Inference in Seoul and Singapore
Amazon Bedrock now offers Claude Opus 5 and Claude Sonnet 5 with in-region inference in Seoul, and Claude Sonnet 5 in Singapore. Requests stay in the selected AWS Region, but throughput is subject to that Region’s capacity and service quota
AI.info Team · Written with AI ·
Local processing comes with a capacity limit
AWS is offering customers in Seoul and Singapore tighter control over where Claude requests run, with a trade-off: when local capacity is constrained, requests cannot spill into another AWS Region. The company announced the availability on September 29, adding Anthropic models to Amazon Bedrock’s in-region inference service in the two markets.
South Korea’s Seoul Region now supports Claude Opus 5 and Claude Sonnet 5. Singapore supports Claude Sonnet 5, but not Opus 5 under this announcement. AWS says prompts and generated responses remain in the Region selected for the request throughout processing. The feature is intended for organizations with strict single-Region data-processing requirements, including customers in financial services, healthcare and the public sector, which AWS names as examples. It describes where Bedrock processes inference; it does not establish that every part of a customer’s application or data workflow automatically stays in that Region.
No routing across Regions
The distinction from cross-Region inference is central. Those profiles route requests across Regions, while the new option sends a request to the selected Region alone. The four AWS blog authors—Aamna Najmi, a senior specialist solutions architect for Generative AI; Alfredo Castillo, a senior specialist solutions architect for Generative AI; Eugenio Soltero, a senior product marketing manager for Amazon Bedrock; and Sofian Hamiti, a technology leader—describe the single-Region boundary:
“Your request is processed entirely within the single AWS Region you specify, and it does not leave that Region.”
Aamna Najmi, Alfredo Castillo, Eugenio Soltero and Sofian Hamiti, AWS blog authors
That control comes with a capacity constraint. AWS says throughput is bounded by the capacity of the Region handling the request, and that each Region’s service quotas apply. Customers choosing strict local processing therefore need to plan around the capacity and quota available in Seoul or Singapore rather than relying on another Region to absorb a spike. Billing follows the standard on-demand pricing for the Region called. Quota consumption, CloudWatch metrics and CloudTrail log entries are also scoped to that Region, with no source-versus-destination distinction in monitoring.
Bedrock keeps familiar API paths
Applications call the models through the bedrock-runtime endpoint using direct model IDs, such as anthropic.claude-opus-5 and anthropic.claude-sonnet-5. AWS lists the Anthropic Messages API and Amazon Bedrock’s InvokeModel and Converse APIs as supported routes. The announcement also says customers can use Bedrock features including Guardrails and intelligent prompt routing.
AWS recommends bedrock-runtime for new applications. Developers can first select a Claude model in the Amazon Bedrock console’s text playground, send prompts, adjust inference parameters and switch between variants without setting up code or an SDK. For programmatic use, AWS’s examples configure the client to the Region where the inference request should be processed: ap-northeast-2 for Seoul and ap-southeast-1 for Singapore.
The post provides examples using Boto3 with InvokeModel in Singapore and Converse in Seoul, as well as an Anthropic SDK example for the Messages API. Its listed prerequisites include an active AWS account with Bedrock access, the AWS CLI, Python 3.8 or later, Boto3, the Anthropic SDK and the Amazon Bedrock Token Generator for authentication. AWS says usage, performance and costs can be monitored through CloudWatch and Cost Explorer.
For teams that need inference requests to stay in one AWS Region, the available choices are specific: Seoul gets Claude Opus 5 and Sonnet 5, while Singapore gets Sonnet 5. Each workload remains subject to the capacity and quota of its selected Region.