World model for autonomous agents: open-source synthetic environment
In brief
- WildAWM generates 341 complex tasks for autonomous agents, each requiring over 100 steps.
- Environments combine graphical interfaces, command lines, and context protocols for flexible agent interaction.
- Two Qwen models trained on synthetic data improve performance across 8 external benchmarks.
Researchers introduce WildAWM, an open-source method to generate training environments for computer-using agents. According to the announcement, the pipeline synthesizes 341 complex tasks requiring over 100 agent steps, with automated checks to ensure difficulty and solvability. Generated environments combine graphical interfaces, context protocols, and command lines, enabling agents to mix visual actions, tool use, and code execution.
The study shows that two Qwen models trained exclusively on this synthetic data improve performance across 8 external benchmarks, including web agents, command-line tasks, and computer-use scenarios. Results indicate that 60% of tasks remain unsolvable even for advanced models like GPT-6-Astra, confirming the environments' challenge level.
This summary gives you the essentials. The full article, its figures and quotes are on Hugging Face Daily Papers.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error