WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution
Open paper
Benchmark
Model
Rank
Results
agents-on-automationbench-public
WorkRover-35B-A3B
–
Success Rate: 28.67
question-answering-on-officeqa-pro-preselected-text
WorkRover-35B-A3B (Claude Code)
–
Accuracy: 48.87
Rank counts only results with a code link.