AI progress is usually judged by what one model does alone, but AI programs that act on their own, called agents, increasingly work in teams with other agents and people. Leaderboards do not reveal whether a model is a good teammate: whether it speaks up, checks others' work, or resists pressure.
The Collaborative Intelligence is a standardized test using real cases with outcomes a program can check. Each model is tested alone, then in a team, and scored only on tasks it can already solve alone, so failures reflect teamwork rather than ability.
Every model tested had clear teamwork weaknesses; the best scored only 82.2. Models with nearly identical general capability differed by up to 11 points as teammates, so solo skill does not guarantee good collaboration.