Tag
This paper presents a multi-agent reasoning framework where multiple foundation models collaborate through structured critique and aggregation, demonstrating that model heterogeneity significantly improves step-wise reasoning accuracy and reduces variance across domains.