Amazon Bedrock now offers single-region inference for Anthropic’s Claude Opus 5 and Sonnet 5 in Seoul, and Claude Sonnet 5 in Singapore. Prompts and generated results stay within the region called for the full request lifecycle, addressing local-processing requirements in fields such as finance, health care and government.
Unlike Bedrock’s cross-region inference profiles, the feature has no routing layer that sends work elsewhere when demand rises. A request made to Seoul or Singapore is served there alone. The trade-off is that throughput is limited by that region’s capacity and its service quotas.
Developers call the models through the Bedrock Runtime endpoint with direct model IDs. Bedrock’s Converse and InvokeModel APIs are supported, as is Anthropic’s Messages API through its SDK. Guardrails and intelligent prompt routing remain available. Billing follows on-demand pricing for the chosen region, while CloudWatch metrics, CloudTrail records and quota use are scoped to the same place.
The release provides a clearer data-residency boundary, not a guarantee of unlimited local capacity. Teams choosing it over cross-region routing need to plan around regional quotas and availability. Current model coverage also differs: Opus 5 is listed for Seoul, while Singapore receives Sonnet 5.