LongCat-2.0

About

LongCat-2.0 is a 1.6 trillion parameter Mixture of Experts model available under an MIT license, featuring approximately 48 billion active parameters and a 1 million token context window. It employs LongCat Sparse Attention and has undergone post-training optimization for coding and agentic tasks. The model was trained using AI ASIC superpods and offers integration with Claude Code, OpenClaw, and Hermes.