agent-as-judge

AJ-Bench introduces a benchmark to evaluate Agent-as-a-Judge systems that interact with environments to verify agent behaviors across 155 tasks in search, data systems, and GUI domains.