Workflow-GYM Benchmarks Computer-Use Agents on Real Long-Horizon Professional Tasks
arXiv (via HuggingFace Daily Papers)·medium signal
Workflow-GYM (arXiv 2606.11042) is a benchmark for long-horizon computer-use agents on real professional-field tasks, aimed at the gap between short tool-use demos and sustained multi-step workplace automation. It gives teams a way to measure whether computer-use agents actually hold up over extended sessions rather than acing isolated steps.