We present a new behavioral evaluation of ChatGPT-3.5 and 4 Large Language Models. Groups of 3 ChatGPT agents were asked to play a collaborative number-guessing game in which each agent had to submit a number, and the sum of the team's numbers needed …