Reproduce “Agent World Model in the Wild: Computer-Use Agent Environment Synthesis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhaoyang Wang, Qianhui Wu, Haorui Wang, Simon Yu, Baolin Peng, Canwen Xu, Hao Cheng, Wenlin Yao and 4 more
WildAWM synthesizes connected personal-computer worlds spanning graphical user interfaces, command lines, and MCP services, then accepts tasks only after execution-based checks. The authors train Qwen3.6-35B-A3B with supervised fine-tuning and reinforcement learning in these worlds. Across eight out-of-distribution benchmarks, its average rises from 42.84 to 53.42; the measures mix pass rates and mean scores, and this does not establish computer-use capability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.