Amazon SageMaker AI Async Inference now supports inline request payloads

This article announces inline payload support for Amazon SageMaker AI Async Inference, allowing customers to send inference payloads directly in API requests without uploading to S3 first.

• New Body parameter accepts payloads up to 128,000 bytes inline in InvokeEndpointAsync API calls
• Eliminates mandatory S3 upload step, reducing latency by one network round-trip per request
• Body and InputLocation parameters are mutually exclusive; API rejects requests setting both
• Simplifies client code by removing S3 bucket provisioning, IAM permissions, and cleanup logic
• Reduces costs by eliminating S3 PUT charges for input uploads on inline invocations
• Available in 31 commercial AWS Regions with no changes needed to endpoint configuration or models
• Backward-compatible; existing InputLocation workflows continue unchanged

Inline payloads are ideal for small JSON prompts and structured data, while S3 uploads remain recommended for payloads exceeding 128KB or when audit trails are required.