Amazon SageMaker AI expanded its model customization capabilities by introducing multi-turn reinforcement learning (RL), enabling models to be fine-tuned on multi-step, agent-based tasks via a serverless infrastructure. This method trains models on complete sequences of decisions within custom agent environments, improving task accuracy while reducing the cost and complexity compared to building bespoke training systems.
The solution integrates with various agent runtimes and manages training workflows end-to-end. It supports models such as Qwen 3.6 27B and GPT-OSS-20B across select AWS regions. While vendor claims emphasize improved task matching to general-purpose models, independent outcomes on model performance or cost savings have yet to be published.
Source & further reading
Amazon SageMaker AI launches multi-turn reinforcement learning for AI agent model customization ↗Source published 2026-06-03. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.