Copertina

LongCat-Flash-Omni Technical Report

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Meituan LongCat Team, Wang, Bairui, Bayan, Xiao, Bin, Zhang, Bo, Rong, Bolin, Chen, Borun, Wan, Chang, Zhang, Chao, Huang, Chen, Chen, Chen, Yang, Chengxu, Yang, Chengzuo, Han, Cong, Peng, Dandan, Ruan, Delian, Xin, Detai, Wang, Disong, Yang, Dongchao, Liu, Fanfan, Chen, Fengjiao, Yang, Fengyu, Dong, Gan, Huang, Gang, Xu, Gang, Wan, Guanglu, Tan, Guoqiang, Yu, Guoqiao, Qiu, Haibo, Lu, Hao, Liu, Hongbo, Xiang, Hongyu, Wu, Jiaheng, Yang, Jian, Liu, Jiaxing, Huang, Jing, Wang, Jingang, Ding, Jinrui, Jiang, Juchao, Kuang, Jun, Wang, Jun, Mei, Junhui, Ding, Ke, Zhang, Kefeng, Chen, Lei, Shi, Liang, Qiao, Limeng, Zheng, Liming, Ma, Lin, Guo, Liuyang, Ma, Liya, Sun, Luying, Gao, Man, Zhu, Mengshen, Cao, Miao, Lin, Minliang, Xu, Nuo, Shi, Peng, Zhang, Qi, Fang, Qian, Wang, Qian, Yang, Qian, Wang, Quanxiu, Weng, Rongxiang, Guo, Rongxin, Liang, Ruoxuan, Yang, Senbin, Xu, Shanbo, Lei, Shanglin, Ye, Shengze, Chen, Shimin, Chen, Shuaiqi, Hu, Shujie, Li, Shuo, Yang, Siqi, Xu, Siyu, Ren, Siyu, Li, Song, Liu, Songxiang, Bai, Tianhao, Dai, Tianye, Hong, Wei, Wang, Wei, Zhao, Weixiao, Cao, Wengang, Zhu, Wenlong, He, Wenlong, Su, Xi, Nan, Xi, Zhao, Xiaohan, Wang, Xiaohao, Zhao, Xiaoyu, Wang, Xiaoyu, Li, Xiaoyu, Pan, Xin, Chen, Xin, Sun, Xiusong, Xiang, Xu, Xing, Xudong, Cao, Xuezhi, Cai, Xunliang, Yang, Yang, Tan, Yanli, Yao, Yao, Sun, Yerui, Chen, Yi, Lu, Yifan, Gong, Yin, Zhang, Yining, Chen, Yitian, Gan, Yiyang, Tang, Yuchen, Xie, Yuchen, Wang, Yueqian, Zheng, Yuewen, Zhang, Yufei, Zhong, Yufeng, Qian, Yulei, Peng, Yuqi, Li, Yuqian, Jiang, Yuwei, Hu, Zeyang, Zhang, Zheng, Tian, Zhengkun, Hong, Zhiqing, Zeng, Zhixiong, Mi, Zhuqi, Li, Ziran, Wang, Ziwen, Zhao, Ziyi, Zhuang, Ziyuan, Zhao, Zizhe
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Multimedia Artificial Intelligence Computation and Language Distributed, Parallel, and Cluster Computing Machine Learning Sound
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

_version_	1866917109667201024
author	Meituan LongCat Team Wang, Bairui Bayan Xiao, Bin Zhang, Bo Rong, Bolin Chen, Borun Wan, Chang Zhang, Chao Huang, Chen Chen, Chen Chen, Chen Yang, Chengxu Yang, Chengzuo Han, Cong Peng, Dandan Ruan, Delian Xin, Detai Wang, Disong Yang, Dongchao Liu, Fanfan Chen, Fengjiao Yang, Fengyu Dong, Gan Huang, Gang Xu, Gang Wan, Guanglu Tan, Guoqiang Yu, Guoqiao Qiu, Haibo Lu, Hao Liu, Hongbo Xiang, Hongyu Wu, Jiaheng Yang, Jian Liu, Jiaxing Huang, Jing Wang, Jingang Ding, Jinrui Jiang, Juchao Kuang, Jun Wang, Jun Mei, Junhui Ding, Ke Zhang, Kefeng Chen, Lei Shi, Liang Qiao, Limeng Zheng, Liming Ma, Lin Guo, Liuyang Ma, Liya Sun, Luying Gao, Man Zhu, Mengshen Cao, Miao Lin, Minliang Xu, Nuo Shi, Peng Zhang, Qi Fang, Qian Wang, Qian Yang, Qian Wang, Quanxiu Weng, Rongxiang Guo, Rongxin Liang, Ruoxuan Yang, Senbin Xu, Shanbo Lei, Shanglin Ye, Shengze Chen, Shimin Chen, Shuaiqi Hu, Shujie Li, Shuo Yang, Siqi Xu, Siyu Ren, Siyu Li, Song Liu, Songxiang Bai, Tianhao Dai, Tianye Hong, Wei Wang, Wei Zhao, Weixiao Cao, Wengang Zhu, Wenlong He, Wenlong Su, Xi Nan, Xi Zhao, Xiaohan Wang, Xiaohao Zhao, Xiaoyu Wang, Xiaoyu Li, Xiaoyu Pan, Xin Chen, Xin Sun, Xiusong Xiang, Xu Xing, Xudong Cao, Xuezhi Cai, Xunliang Yang, Yang Tan, Yanli Yao, Yao Sun, Yerui Chen, Yi Lu, Yifan Gong, Yin Zhang, Yining Chen, Yitian Gan, Yiyang Tang, Yuchen Xie, Yuchen Wang, Yueqian Zheng, Yuewen Zhang, Yufei Zhong, Yufeng Qian, Yulei Peng, Yuqi Li, Yuqian Jiang, Yuwei Hu, Zeyang Zhang, Zheng Tian, Zhengkun Hong, Zhiqing Zeng, Zhixiong Mi, Zhuqi Li, Ziran Wang, Ziwen Zhao, Ziyi Zhuang, Ziyuan Zhao, Zizhe
author_facet	Meituan LongCat Team Wang, Bairui Bayan Xiao, Bin Zhang, Bo Rong, Bolin Chen, Borun Wan, Chang Zhang, Chao Huang, Chen Chen, Chen Chen, Chen Yang, Chengxu Yang, Chengzuo Han, Cong Peng, Dandan Ruan, Delian Xin, Detai Wang, Disong Yang, Dongchao Liu, Fanfan Chen, Fengjiao Yang, Fengyu Dong, Gan Huang, Gang Xu, Gang Wan, Guanglu Tan, Guoqiang Yu, Guoqiao Qiu, Haibo Lu, Hao Liu, Hongbo Xiang, Hongyu Wu, Jiaheng Yang, Jian Liu, Jiaxing Huang, Jing Wang, Jingang Ding, Jinrui Jiang, Juchao Kuang, Jun Wang, Jun Mei, Junhui Ding, Ke Zhang, Kefeng Chen, Lei Shi, Liang Qiao, Limeng Zheng, Liming Ma, Lin Guo, Liuyang Ma, Liya Sun, Luying Gao, Man Zhu, Mengshen Cao, Miao Lin, Minliang Xu, Nuo Shi, Peng Zhang, Qi Fang, Qian Wang, Qian Yang, Qian Wang, Quanxiu Weng, Rongxiang Guo, Rongxin Liang, Ruoxuan Yang, Senbin Xu, Shanbo Lei, Shanglin Ye, Shengze Chen, Shimin Chen, Shuaiqi Hu, Shujie Li, Shuo Yang, Siqi Xu, Siyu Ren, Siyu Li, Song Liu, Songxiang Bai, Tianhao Dai, Tianye Hong, Wei Wang, Wei Zhao, Weixiao Cao, Wengang Zhu, Wenlong He, Wenlong Su, Xi Nan, Xi Zhao, Xiaohan Wang, Xiaohao Zhao, Xiaoyu Wang, Xiaoyu Li, Xiaoyu Pan, Xin Chen, Xin Sun, Xiusong Xiang, Xu Xing, Xudong Cao, Xuezhi Cai, Xunliang Yang, Yang Tan, Yanli Yao, Yao Sun, Yerui Chen, Yi Lu, Yifan Gong, Yin Zhang, Yining Chen, Yitian Gan, Yiyang Tang, Yuchen Xie, Yuchen Wang, Yueqian Zheng, Yuewen Zhang, Yufei Zhong, Yufeng Qian, Yulei Peng, Yuqi Li, Yuqian Jiang, Yuwei Hu, Zeyang Zhang, Zheng Tian, Zhengkun Hong, Zhiqing Zeng, Zhixiong Mi, Zhuqi Li, Ziran Wang, Ziwen Zhao, Ziyi Zhuang, Ziyuan Zhao, Zizhe
contents	We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.
format	Preprint
id	arxiv_https___arxiv_org_abs_2511_00279
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	LongCat-Flash-Omni Technical Report Meituan LongCat Team Wang, Bairui Bayan Xiao, Bin Zhang, Bo Rong, Bolin Chen, Borun Wan, Chang Zhang, Chao Huang, Chen Chen, Chen Chen, Chen Yang, Chengxu Yang, Chengzuo Han, Cong Peng, Dandan Ruan, Delian Xin, Detai Wang, Disong Yang, Dongchao Liu, Fanfan Chen, Fengjiao Yang, Fengyu Dong, Gan Huang, Gang Xu, Gang Wan, Guanglu Tan, Guoqiang Yu, Guoqiao Qiu, Haibo Lu, Hao Liu, Hongbo Xiang, Hongyu Wu, Jiaheng Yang, Jian Liu, Jiaxing Huang, Jing Wang, Jingang Ding, Jinrui Jiang, Juchao Kuang, Jun Wang, Jun Mei, Junhui Ding, Ke Zhang, Kefeng Chen, Lei Shi, Liang Qiao, Limeng Zheng, Liming Ma, Lin Guo, Liuyang Ma, Liya Sun, Luying Gao, Man Zhu, Mengshen Cao, Miao Lin, Minliang Xu, Nuo Shi, Peng Zhang, Qi Fang, Qian Wang, Qian Yang, Qian Wang, Quanxiu Weng, Rongxiang Guo, Rongxin Liang, Ruoxuan Yang, Senbin Xu, Shanbo Lei, Shanglin Ye, Shengze Chen, Shimin Chen, Shuaiqi Hu, Shujie Li, Shuo Yang, Siqi Xu, Siyu Ren, Siyu Li, Song Liu, Songxiang Bai, Tianhao Dai, Tianye Hong, Wei Wang, Wei Zhao, Weixiao Cao, Wengang Zhu, Wenlong He, Wenlong Su, Xi Nan, Xi Zhao, Xiaohan Wang, Xiaohao Zhao, Xiaoyu Wang, Xiaoyu Li, Xiaoyu Pan, Xin Chen, Xin Sun, Xiusong Xiang, Xu Xing, Xudong Cao, Xuezhi Cai, Xunliang Yang, Yang Tan, Yanli Yao, Yao Sun, Yerui Chen, Yi Lu, Yifan Gong, Yin Zhang, Yining Chen, Yitian Gan, Yiyang Tang, Yuchen Xie, Yuchen Wang, Yueqian Zheng, Yuewen Zhang, Yufei Zhong, Yufeng Qian, Yulei Peng, Yuqi Li, Yuqian Jiang, Yuwei Hu, Zeyang Zhang, Zheng Tian, Zhengkun Hong, Zhiqing Zeng, Zhixiong Mi, Zhuqi Li, Ziran Wang, Ziwen Zhao, Ziyi Zhuang, Ziyuan Zhao, Zizhe Multimedia Artificial Intelligence Computation and Language Distributed, Parallel, and Cluster Computing Machine Learning Sound We introduce LongCat-Flash-Omni, a state-of-the-art open-source omni-modal model with 560 billion parameters, excelling at real-time audio-visual interaction. By adopting a curriculum-inspired progressive training strategy that transitions from simpler to increasingly complex modality sequence modeling tasks, LongCat-Flash-Omni attains comprehensive multimodal capabilities while maintaining strong unimodal capability. Building upon LongCat-Flash, which adopts a high-performance Shortcut-connected Mixture-of-Experts (MoE) architecture with zero-computation experts, LongCat-Flash-Omni integrates efficient multimodal perception and speech reconstruction modules. Despite its immense size of 560B parameters (with 27B activated), LongCat-Flash-Omni achieves low-latency real-time audio-visual interaction. For training infrastructure, we developed a modality-decoupled parallelism scheme specifically designed to manage the data and model heterogeneity inherent in large-scale multimodal training. This innovative approach demonstrates exceptional efficiency by sustaining over 90% of the throughput achieved by text-only training. Extensive evaluations show that LongCat-Flash-Omni achieves state-of-the-art performance on omni-modal benchmarks among open-source models. Furthermore, it delivers highly competitive results across a wide range of modality-specific tasks, including text, image, and video understanding, as well as audio understanding and generation. We provide a comprehensive overview of the model architecture design, training procedures, and data strategies, and open-source the model to foster future research and development in the community.
title	LongCat-Flash-Omni Technical Report
topic	Multimedia Artificial Intelligence Computation and Language Distributed, Parallel, and Cluster Computing Machine Learning Sound
url	https://arxiv.org/abs/2511.00279