Meituan Releases LongCat-2.0: A 1.6T-Parameter Open MoE Model with Native 1M Context and LongCat Sparse Attention

Short summary

Meituan released LongCat-2.0, a 1.6-trillion-parameter mixture-of-experts model with native 1-million-token context using sparse attention. The model activates ~48 billion parameters per token for efficiency. It trains and serves on domestic ASIC superpods as a competitive open-source alternative.

• •1.6T parameter MoE model with 1M token context window
• •48B active parameters per token via LongCat Sparse Attention
• •End-to-end training and serving on domestic AI ASIC hardware

Generated with AI, which can make mistakes.