About
LongCat is an open AI model family created and published by Chinese giant Meituan (LongCat team) since September 2025. Family strategy: open-source-first, with every model released under the MIT licence on HuggingFace under the meituan-longcat organisation. The line spans several branches: (1) MoE LLMs — LongCat-Flash-Chat (September 2025, the first public release), LongCat-Flash-Thinking (CoT), LongCat-Flash-Lite (smaller, FP8), LongCat-Flash-Prover (formal mathematics, 561B); (2) multimodal — LongCat-Flash-Omni (any-to-any), LongCat-Image (6B), LongCat-Image-Edit-Turbo, LongCat-Video (October 2025), LongCat-Video-Avatar-1.5 (talking-head); (3) audio — LongCat-AudioDiT-1B, LongCat-AudioDiT-3.5B; (4) any-to-any — LongCat-Next (74B); (5) frontier — LongCat-2.0 (December 2025, 1.6T MoE parameters, ~48B active, fully trained on Chinese AI ASIC superpods on 35T+ tokens). All models power the Xiaomei agent in Meituan products (food delivery, hotels).
