RpR (RolePlay with Reasoning) models which are built on RPMax datasets with properly trained multi-turn reasoning.