Research
Agent World Model in the Wild proposes scalable computer-use agent environment synthesis
WildAWM synthesizes 341 long-horizon environment-task pairs spanning GUI, MCP and command-line interfaces, built with open-source models. Two Qwen models trained only on these synthetic worlds show gains across 8 out-of-distribution benchmarks, and GPT-6-Astra solves only 60% of the tasks.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
Agent World Model in the Wild: Computer-Use Agent Environment Synthesis