Amazon Web Services has launched Amazon SageMaker AI Training Sessions, a serverless feature designed for customizing AI models. The tool allows developers to manage training logic while AWS handles GPU infrastructure, scaling, and distributed execution.
The feature supports methods such as supervised fine-tuning, reinforcement learning, custom loss functions, and multi-turn reinforcement learning. Developers can access these capabilities via Python SDK clients from local machines or SageMaker Studio.
Training Sessions supports open-weight models including Llama, Gemma, Qwen, Nemotron, and gpt-oss. At launch, the service is available in the US East (N. Virginia), US West (Oregon), and Asia Pacific (Mumbai) regions.
Security is managed through IAM policies and automatic encryption at rest in S3 with optional customer-managed KMS keys. Pricing is based on the volume of tokens processed during training sessions.
This release follows previous serverless updates, including full fine-tuning introduced in August 2026 and custom sequence length support added on October 21, 2026.
