Alibaba Qwen introduces QwenGyre, an elastic reinforcement learning framework for long-horizon agents
Alibaba's Qwen team unveiled QwenGyre, an elastic reinforcement learning framework for training agents on tasks that run for hours and process close to 1 million tokens. The system shifts idle GPUs between rollout and training and turns branching agent histories into bounded training samples. Tested on Qwen 3.6 122B and Qwen 3.8 2.4T models, it delivers 1.21x to 1.85x speedups over comparable meth
