Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents

Alibaba’s Qwen team has released QwenGyre, an open-source reinforcement learning framework designed to train AI agents capable of performing tasks over extended durations. The system utilizes black-box harnesses to allow agents to execute complex, multi-hour processes like coding without requiring internal modifications to the software environment. By facilitating long-horizon training, this framework aims to improve the performance of autonomous agents in workflows that require sustained interaction and decision-making over long timeframes.