Back to News
Qwen team at Alibaba•June 24, 2026
Qwen-AgentWorld
Open Source
open-source
language model
simulation
AI
### TL;DR
Qwen-AgentWorld is an open-source language world model developed by QwenLM, designed to simulate agentic environments across seven unified domains: MCP, Search, Terminal, SWE, Android, Web, and OS. It employs a three-stage training pipeline—CPT, SFT, and RL—on over 10 million real-world interaction trajectories to enhance its simulation fidelity and generalization capabilities.
Key Insights & Metrics
Pricing
Free
Cost structure
Version
Initial Release
Current release version
Hardware
To run the 35B model efficiently, you will need a multi-GPU setup; ideally 4 GPUs with at least 70GB to 80GB of total VRAM
Compute requirements
Category
Open Source
Licensing model
Region
China
Primary region
Key Features
- Simulates agentic environments across seven unified domains
- Employs a three-stage training pipeline: CPT, SFT, and RL
- Trained on over 10 million real-world interaction trajectories
- Available on Hugging Face and ModelScope
- Technical report accessible on arXiv
Discussion
0
Upvotes
0
Downvotes
0 reviews
Sign in to leave a review
Reviews
No reviews yet. Be the first to review!