Batch processing throughput improvement

LongCat-Flash-Thinking: MoE and asynchronous RL support, AIME25 saves more tokens

LongCat-Flash-Thinking combines AI with MoE, asynchronous RL, and agent-native tools, achieving state-of-the-art performance in logic, math, coding, a...