Can LLM Teams Play What? Where? When?

In three linesStudy on LLM teams playing ChGK (collective reasoning quiz). Three strategies tested: Voting, Silent Team (captain observes answers), Talkative Team (captain observes answers + rationales). On 572 questions from 2025, teams outperform single models (+20 points). Best team: 44.23% accuracy, approaching human performance. Sharing rationales mitigates errors.

Summary generated by Claude — human-verified