Qwen-AgentWorld: Language World Models for General Agents across Seven Environments
Alibaba/Qwen
Alibaba's Qwen team published Qwen-AgentWorld (arXiv 2606.24597, June 23), introducing language world models — 35B-A3B and 397B-A17B MoE variants — that simulate seven agentic environments: MCP, Search, Terminal, Software Engineering, Android, Web, and OS. Trained on over 10 million real environment interaction trajectories. Also introduces AgentWorldBench covering all seven domains. The models can serve as scalable RL training simulators or as warm-up training for downstream agent tasks.
Why it matters
The first language world model operating at this breadth of agentic environments — providing a unified simulator for RL training across seven domains rather than requiring seven separate real-world environments — could meaningfully reduce the cost and friction of training capable agents. Top-voted paper on HF Daily Papers for June 24 (36 upvotes).
Importance: 3/5
First language world model spanning seven agentic environments; top HF Daily Papers for June 24 (36 upvotes); enables scalable synthetic RL environment simulation