Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.27538 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910082337341440 |
|---|---|
| author | Meituan LongCat Team Xiao, Bin Wang, Chao Li, Chengjiang Zhang, Chi Peng, Chong Yu, Hang Yang, Hao Yan, Haonan Sun, Haoze Zhao, Haozhe Liu, Hong Su, Hui Zhang, Jiaqi Wang, Jiawei Li, Jing Zhang, Kefeng Zhang, Manyuan Jing, Minhao Pei, Peng Chen, Quan Xue, Taofeng Pan, Tongxin Li, Xiaotong Li, Xiaoyang Zhao, Xiaoyu Hu, Xing Lin, Xinyang Cai, Xunliang Bai, Yan Feng, Yan Li, Yanjie Qiu, Yao Sun, Yerui Lu, Yifan Luo, Ying Mei, Yipeng Chen, Yitian Xie, Yuchen Liu, Yufang Chen, Yufei Qian, Yulei Peng, Yuqi Yu, Zhihang Han, Zhixiong Wang, Changran Chen, Chen Zheng, Dian Chen, Fengjiao Yang, Ge Guo, Haowei Wang, Haozhe Li, Hongyu Jiang, Huicheng Hong, Jiale Zou, Jialv Li, Jiamu Lin, Jianping Liu, Jiaxing Yang, Jie Jin, Jing Kuang, Jun She, Juncheng Luo, Kunming Gao, Kuofeng Qiu, Lin Guo, Linsen Huang, Mianqiu Li, Qi Wang, Qian Li, Rumei Ren, Siyu Wang, Wei He, Wenlong Chen, Xi Liu, Xiao Li, Xiaoyu Huang, Xu Zhu, Xuanyu Cao, Xuezhi Zhu, Yaoming Cao, Yifei Jia, Yimeng Jiang, Yizhen Gao, Yufei Hu, Zeyang Yuan, Zhenlong Zhang, Zijian Wang, Ziwen |
| author_facet | Meituan LongCat Team Xiao, Bin Wang, Chao Li, Chengjiang Zhang, Chi Peng, Chong Yu, Hang Yang, Hao Yan, Haonan Sun, Haoze Zhao, Haozhe Liu, Hong Su, Hui Zhang, Jiaqi Wang, Jiawei Li, Jing Zhang, Kefeng Zhang, Manyuan Jing, Minhao Pei, Peng Chen, Quan Xue, Taofeng Pan, Tongxin Li, Xiaotong Li, Xiaoyang Zhao, Xiaoyu Hu, Xing Lin, Xinyang Cai, Xunliang Bai, Yan Feng, Yan Li, Yanjie Qiu, Yao Sun, Yerui Lu, Yifan Luo, Ying Mei, Yipeng Chen, Yitian Xie, Yuchen Liu, Yufang Chen, Yufei Qian, Yulei Peng, Yuqi Yu, Zhihang Han, Zhixiong Wang, Changran Chen, Chen Zheng, Dian Chen, Fengjiao Yang, Ge Guo, Haowei Wang, Haozhe Li, Hongyu Jiang, Huicheng Hong, Jiale Zou, Jialv Li, Jiamu Lin, Jianping Liu, Jiaxing Yang, Jie Jin, Jing Kuang, Jun She, Juncheng Luo, Kunming Gao, Kuofeng Qiu, Lin Guo, Linsen Huang, Mianqiu Li, Qi Wang, Qian Li, Rumei Ren, Siyu Wang, Wei He, Wenlong Chen, Xi Liu, Xiao Li, Xiaoyu Huang, Xu Zhu, Xuanyu Cao, Xuezhi Zhu, Yaoming Cao, Yifei Jia, Yimeng Jiang, Yizhen Gao, Yufei Hu, Zeyang Yuan, Zhenlong Zhang, Zijian Wang, Ziwen |
| contents | The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_27538 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LongCat-Next: Lexicalizing Modalities as Discrete Tokens Meituan LongCat Team Xiao, Bin Wang, Chao Li, Chengjiang Zhang, Chi Peng, Chong Yu, Hang Yang, Hao Yan, Haonan Sun, Haoze Zhao, Haozhe Liu, Hong Su, Hui Zhang, Jiaqi Wang, Jiawei Li, Jing Zhang, Kefeng Zhang, Manyuan Jing, Minhao Pei, Peng Chen, Quan Xue, Taofeng Pan, Tongxin Li, Xiaotong Li, Xiaoyang Zhao, Xiaoyu Hu, Xing Lin, Xinyang Cai, Xunliang Bai, Yan Feng, Yan Li, Yanjie Qiu, Yao Sun, Yerui Lu, Yifan Luo, Ying Mei, Yipeng Chen, Yitian Xie, Yuchen Liu, Yufang Chen, Yufei Qian, Yulei Peng, Yuqi Yu, Zhihang Han, Zhixiong Wang, Changran Chen, Chen Zheng, Dian Chen, Fengjiao Yang, Ge Guo, Haowei Wang, Haozhe Li, Hongyu Jiang, Huicheng Hong, Jiale Zou, Jialv Li, Jiamu Lin, Jianping Liu, Jiaxing Yang, Jie Jin, Jing Kuang, Jun She, Juncheng Luo, Kunming Gao, Kuofeng Qiu, Lin Guo, Linsen Huang, Mianqiu Li, Qi Wang, Qian Li, Rumei Ren, Siyu Wang, Wei He, Wenlong Chen, Xi Liu, Xiao Li, Xiaoyu Huang, Xu Zhu, Xuanyu Cao, Xuezhi Zhu, Yaoming Cao, Yifei Jia, Yimeng Jiang, Yizhen Gao, Yufei Hu, Zeyang Yuan, Zhenlong Zhang, Zijian Wang, Ziwen Computer Vision and Pattern Recognition Computation and Language The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next |
| title | LongCat-Next: Lexicalizing Modalities as Discrete Tokens |
| topic | Computer Vision and Pattern Recognition Computation and Language |
| url | https://arxiv.org/abs/2603.27538 |