LongCat-Flash-Chat

LongCat-Flash-Chat is Meituan's first open-source foundation model, a 560B parameter Mixture-of-Experts (MoE) model that dynamically activates 18.6B-31.3B parameters (~27B average) based on contextual demands. It features Zero-Computation Experts for efficient routing and supports 128K context. Optimized for conversational and agentic tasks, it shows competitive performance across reasoning, coding, instruction following, and domain benchmarks with particular strengths in tool use and complex multi-step interactions. Achieves over 100 tokens per second on H800 GPUs.