Berkeley Function Calling Leaderboard v4

Lasso Security found that watermarking large language models changes AI agent behavior, with six of seven tested models scoring lower on tool-call tasks and four of those declines statistically signif…