Meituan on July 6 open-sourced its LongCat-2.0 large language model with 1.6 trillion total parameters and an average activation of about 48 billion, according to Jiemian News. The company said the model introduces LongCat sparse attention and N-gram Embedding to improve long-context processing efficiency and token-level representations, and uses dynamic activation to enhance code understanding, generation and execution performance.
The open-source release includes multiple precision versions, including BF16, FP8 and INT8, and Meituan said it will also release inference optimizations tailored for domestic Chinese computing platforms.