1 paper
Jinge Wu, Hongjian Zhou, Mingde Zeng +8
Reproducing and comparing deep research agents today is hard: the same backbone evaluated on the same benchmark can report different accuracies across papers because the harness an…