# Alibaba launches QwenGyre framework for ultra‑long‑horizon RL training

> Alibaba’s Qwen team unveiled QwenGyre, an elastic reinforcement‑learning system that can train coding agents over hour‑long rollouts handling up to 1 million tokens.

Oossa · 2026-10-12 · https://oossa.com/en/alibaba-launches-qwengyre-framework-for-ultra-long-horizon-rl-training

Alibaba’s Qwen team announced a new framework called QwenGyre. It is an elastic reinforcement‑learning (RL) system designed for training agents that need to run for many hours. QwenGyre can handle rollouts that process close to 1 million tokens without changing the code of the agents. The system is aimed at coding agents that learn through black‑box test harnesses.

## The facts

- QwenGyre was announced on 2026-10-12 (“Mon Oct 12 2026”).
- It targets rollouts lasting hours and processing about 1 million tokens.

## Why it matters

The framework could let developers build more capable coding assistants that learn from very long, realistic usage sessions.

## Sources & references

1. [Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents](https://www.marktechpost.com/2026/10/12/alibaba-qwen-introduces-qwengyre-elastic-reinforcement-learning-rl-for-xlong-horizon-agents/) – MarkTechPost, 2026-10-12

Last updated: 2026-10-12