Recovering Policy-Induced Errors: Benchmarking and Trajectory Synthesis for Robust GUI Agents
Open paper
Benchmark
Model
Rank
Results
computer-use-agents-on-windowsagentarena-15-steps
RoTS-32B
#1
Success Rate: 35.9
computer-use-agents-on-windowsagentarena-50-steps
RoTS-32B
#7
Success Rate: 39.1
Rank counts only results with a code link.