Researchers evaluated 8 model selection strategies for multi-agent systems, finding that expanding candidate pool sizes can degrade performance and introducing arbitrary models can cause system instability. The best strategy is selecting candidates within a single model family.