Meituan has recently open-sourced its proprietary 560 billion parameter hybrid expert model, LongCat-Flash. Built upon the Mixture of Experts (MoE) architecture, this model boasts support for dynamic computation, significantly lowering the bar for enterprise-level adoption. It is now available on both the Hugging Face and GitHub platforms.