# Alibaba Opens QwenGyre for Long-Running AI Agents

> Alibaba's open-source QwenGyre framework trains AI agents to run for hours without breaking.
- Title: Alibaba Opens QwenGyre for Long-Running AI Agents | Shortify
- Summary: Alibaba's open-source QwenGyre framework trains AI agents to run for hours without breaking. Alibaba released QwenGyre, an open-source reinforcement learning…
- Keywords: alibaba, qwengyre, reinforcement learning, ai agents, training time, coding tasks, open source, alibaba qwen, ai framework, token usage, startups, business
- Source: Startup Fortune — https://startupfortune.com/alibabas-qwen-team-open-sources-qwengyre-to-train-ai-agents-that-run-for-hours
- Published: 2026-10-13T01:20:21+00:00
- Read time: 1 min
- Topics: alibaba, qwengyre, reinforcement learning, ai agents, training time, coding tasks, open source, alibaba qwen, ai framework, token usage, startups, business
## Alibaba Releases QwenGyre Framework
Alibaba's Qwen team published QwenGyre, a reinforcement learning framework for training AI agents that run for hours.
## QwenGyre Targets Long Tasks
The framework trains coding agents through unmodified black-box harnesses, targeting rollouts that run for hours and approach 1 million tokens.
## Training Time Reduced by 44%
QwenGyre cut training time by 44%, completing 48 training steps in 75.42 hours instead of 134.55.
## Improved Coding Task Pass Rate
On NL2RepoBench, the model's pass rate climbed from 52.48% to 58.54%.
## Elastic Scheduling Fix
QwenGyre divides GPU capacity into compute units to share hardware dynamically during training.
## Alibaba's 10 Trillion Parameter Plan
Alibaba CEO Eddie Wu plans to train an AI model with 5 to 10 trillion parameters by 2032.
## Open Infrastructure for Teams
QwenGyre trains through existing agent harnesses without requiring re-architecting.
## QwenGyre Adoption Uncertain
Whether QwenGyre gets adopted outside Alibaba remains an open question.
## Key takeaway

Alibaba released QwenGyre, an open-source reinforcement learning framework that trains AI agents to run for hours, cutting training time by 44% and improving coding task pass rates.