CooperBench is the first benchmark for evaluating AI agent teams, agents as teammates, and human-AI collaboration. Over 600 collaborative coding tasks reveal that agent teams achieve 30% lower success rates when working together compared to performing tasks individually.