Meituan has open-sourced LongCat-2.0, simultaneously releasing inference code for domestically produced SIM cards.

Meituan has officially open-sourced its trillion-parameter large-scale model, LongCat-2.0, with a total of 1.6T parameters and an average activation of 48B, specifically designed for agentic coding tasks. The model innovatively introduces LongCat sparse attention and N-gram embedding to improve long context processing and code execution performance, and has been successfully validated for inference on a domestic computing cluster with 50,000 GPUs. This open-source release also includes BF16/FP8/INT8 multi-precision versions and optimized inference code for domestic chips.