New AI Training Method Balances Reasoning Power and Computational Cost

LongCat-Flash-Thinking-2601 combines multiple pre-training strategies with scaled reinforcement learning to build AI models that reason effectively without wasting computational resources.