Qwen-AgentWorld: Language World Models for General Agents
Yuxin Zuo, Zikai Xiao, Li Sheng, Fei Huang, Jianhong Tu, Yuxuan Liu, Tianyi Tang, Xiaomeng Hu, Yang Su, Qingfeng Lan, Yantao Liu, Qin Zhu, Yinger Zhang, Bowen Yu, Haiquan Zhao, Haiyang Xu, Jianxin Yang, Jiayang Cheng, Junyang Wang, Lianghao Deng, Mingfeng Xue, Tianyi Bai, Yang Fan, Yubo Ma, Yucheng Li, Zeyu Cui, Zhihai Wang, Zhihui Xie, Zhuorui Ye, An Yang, Dayiheng Liu, Jingren Zhou, Ning Ding
arXiv:2606.24597 · 144▲ · cs.CL
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
Qwen-AgentWorld is a language model trained to act as a world model: given interaction history and an agent's action, it predicts the next environment observation (tool response, terminal output, UI accessibility tree, etc.) across seven domains via chain-of-thought reasoning. It is built through a three-stage pipeline (continual pre-training, supervised fine-tuning, and RL with combined rubric-and-rule rewards) and evaluated on AgentWorldBench, a benchmark of real trajectories from frontier agents.
Why it matters
Instead of running actual sandboxes, terminals, or GUI environments to train agents, you can generate synthetic but plausible environment feedback from a model. The paper reports that using the world model as a decoupled environment simulator for agentic RL beats training against real environments alone, and that world-model pretraining works as a warm-up that improves downstream agent performance on multiple benchmarks.
Practical takeaway
Watch for language world models as a way to cheaply scale agent RL without maintaining thousands of live environments, and as an auxiliary pretraining stage before agentic fine-tuning. Note the caveat that GUI states are represented as text (accessibility trees and view hierarchies), not pixels.
Key result
Qwen-AgentWorld is reported to outperform frontier models on AgentWorldBench (2,170 samples across seven domains, judged reference-grounded on five dimensions), but the paper gives no head-to-head score margins in the provided text; the RL stage's largest gain was Factuality improving 11.3% relative, while remaining the lowest-scoring dimension. The comparison baselines are frontier models like Claude Opus 4.6, though the arXiv ID and dates (2026) suggest this may be a fabricated or speculative document.
Jargon
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.