List your company

Multi-Agent Collaboration Role Overstepping on SWE hard subset (dev)

0Overstepping Rate (<INFO>)ChatDev
-1.6649.56820.832.032Apr 3, 2026

Evaluation results

MethodLinks
ChatDevCEO Configuration=FT, CPO Configuration=FT, Backbone=Llama, System=ChatDev, Role Clarity Regularization=enabled2026.0400-41.6-42.8
ChatDevCEO Configuration=FT, CPO Configuration=Base, Backbone=Llama, System=ChatDev, Role Clarity Regularization=enabled2026.040.50.5-41.1-42.3
ChatDevCEO Configuration=Base, CPO Configuration=FT, Backbone=Llama, System=ChatDev, Role Clarity Regularization=enabled2026.0414.415.6-27.2-27.2
ChatDevCEO Configuration=Base, CPO Configuration=Base, Backbone=Llama, System=ChatDev, Role Clarity Regularization=enabled2026.0441.642.8——