LRM has developed a powerful CoT reasoning ability through a simple yet effective RLVR paradigm. However, the lengthy output associated with it significantly increases reasoning costs and impacts ...