AWS Adds Optimized Deployments for Foundation Models in SageMaker JumpStart

Amazon Web Services announced in a press release that SageMaker JumpStart now supports optimized deployments for foundation models. The feature enables users to deploy models with pre-configured settings designed for specific use cases and performance constraints, such as content generation, summarization, or question answering.

The new capability includes task-aware configurations that allow users to optimize for cost, throughput, latency, or balanced performance. More than 30 models are supported, including Meta Llama 3.1 and 3.2, Microsoft Phi-3, Mistral AI’s Mistral-Small-24B-Instruct-2501, Qwen 2 and 3 series, Google Gemma, and TII Falcon3. Users can view metrics such as P50 latency, time to first token, and throughput before deployment.

Models can be deployed to SageMaker AI Managed Inference endpoints or SageMaker HyperPod clusters using pre-set configurations that maintain full visibility into deployment details. All deployments use SageMaker’s VPC capabilities for data control and enterprise security. The feature is available in all regions where SageMaker JumpStart is supported.