What changed
Amazon SageMaker Inference has introduced several new features and updates as part of its 2026 launch. These include enhanced inference recommendations available in SageMaker AI Studio, the SageMaker Python SDK v3, and support for LoRA adapters. Additionally, the service now offers SLA-based concurrency search and an agent skill for optimized generative AI inference.
What this means in practice
The new features allow teams to better manage their model configurations and optimize performance. Inference recommendations are now available in SageMaker AI Studio as a guided, low-code workflow. Users can select a use-case profile, optimization goal, and model, and the system benchmarks candidate configurations on real GPU infrastructure. The results are ranked based on TTFT, inter-token latency, throughput, and cost, enabling users to deploy the chosen configuration to a real-time endpoint.
The SageMaker Python SDK v3 integrates these recommendations into the same interface used for training and deployment, allowing for seamless workflows in notebooks, CI pipelines, and MLOps processes. LoRA adapter support enables teams to specialize a foundation model for multiple use cases, with the service benchmarking combinations of adapter count and concurrency on real GPU infrastructure.
The SLA-based concurrency search helps users find the highest concurrency their endpoint can sustain without breaching latency or error targets. This feature automates the process of setting concurrency levels, reducing manual effort and improving efficiency.
Limitations and availability
The new features are available in specific regions and through certain tools like SageMaker AI Studio and the SageMaker Python SDK v3. The agent skill for optimized generative AI inference is available through the Agent Toolkit for AWS, supporting coding agents such as Kiro. The supplied announcement does not specify exact availability timelines for all features, and some functionalities may require additional setup or configuration.
