LongCat-2.0 Open

Overview

LongCat-2.0 Open is Meituan’s next-generation open-weight LLM that supports contexts up to 1M tokens, making it suitable for entire codebases or large document corpora.[2] It is trained on 1.6 trillion parameters with a mixture-of-experts architecture and reportedly uses 50,000 domestically made Chinese AI chips, aligning with China’s drive for onshore compute.[2][9] The model targets research, enterprise knowledge bases, and long-form analysis where transparency and local deployment are important.