Imagen de Portada

Kimi K2: Open Agentic Intelligence

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Kimi Team, Bai, Yifan, Bao, Yiping, Charles, Y., Chen, Cheng, Chen, Guanduo, Chen, Haiting, Chen, Huarong, Chen, Jiahao, Chen, Ningxin, Chen, Ruijue, Chen, Yanru, Chen, Yuankun, Chen, Yutian, Chen, Zhuofu, Cui, Jialei, Ding, Hao, Dong, Mengnan, Du, Angang, Du, Chenzhuang, Du, Dikang, Du, Yulun, Fan, Yu, Feng, Yichen, Fu, Kelin, Gao, Bofei, Gao, Chenxiao, Gao, Hongcheng, Gao, Peizhong, Gao, Tong, Ge, Yuyao, Geng, Shangyi, Gu, Qizheng, Gu, Xinran, Guan, Longyu, Guo, Haiqing, Guo, Jianhang, Hao, Xiaoru, He, Tianhong, He, Weiran, He, Wenyang, He, Yunjia, Hong, Chao, Hu, Hao, Hu, Yangyang, Hu, Zhenxing, Huang, Weixiao, Huang, Zhiqi, Huang, Zihao, Jiang, Tao, Jiang, Zhejun, Jin, Xinyi, Kang, Yongsheng, Lai, Guokun, Li, Cheng, Li, Fang, Li, Haoyang, Li, Ming, Li, Wentao, Li, Yang, Li, Yanhao, Li, Yiwei, Li, Zhaowei, Li, Zheming, Lin, Hongzhan, Lin, Xiaohan, Lin, Zongyu, Liu, Chengyin, Liu, Chenyu, Liu, Hongzhang, Liu, Jingyuan, Liu, Junqi, Liu, Liang, Liu, Shaowei, Liu, T. Y., Liu, Tianwei, Liu, Weizhou, Liu, Yangyang, Liu, Yibo, Liu, Yiping, Liu, Yue, Liu, Zhengying, Lu, Enzhe, Lu, Haoyu, Lu, Lijun, Luo, Yashuo, Ma, Shengling, Ma, Xinyu, Ma, Yingwei, Mao, Shaoguang, Mei, Jie, Men, Xin, Miao, Yibo, Pan, Siyuan, Peng, Yebo, Qin, Ruoyu, Qin, Zeyu, Qu, Bowen, Shang, Zeyu, Shi, Lidong, Shi, Shengyuan, Song, Feifan, Su, Jianlin, Su, Zhengyuan, Sui, Lin, Sun, Xinjie, Sung, Flood, Tai, Yunpeng, Tang, Heyi, Tao, Jiawen, Teng, Qifeng, Tian, Chaoran, Wang, Chensi, Wang, Dinglu, Wang, Feng, Wang, Hailong, Wang, Haiming, Wang, Jianzhou, Wang, Jiaxing, Wang, Jinhong, Wang, Shengjie, Wang, Shuyi, Wang, Si, Wang, Xinyuan, Wang, Yao, Wang, Yejie, Wang, Yiqin, Wang, Yuxin, Wang, Yuzhi, Wang, Zhaoji, Wang, Zhengtao, Wang, Zhexu, Wei, Chu, Wei, Qianqian, Wu, Haoning, Wu, Wenhao, Wu, Xingzhe, Wu, Yuxin, Xiao, Chenjun, Xie, Jin, Xie, Xiaotong, Xiong, Weimin, Xu, Boyu, Xu, Jinjing, Xu, L. H., Xu, Lin, Xu, Suting, Xu, Weixin, Xu, Xinran, Xu, Yangchuan, Xu, Ziyao, Xu, Jing, Yan, Junjie, Yan, Yuzi, Yang, Hao, Yang, Xiaofei, Yang, Yi, Yang, Ying, Yang, Zhen, Yang, Zhilin, Yang, Zonghan, Yao, Haotian, Yao, Xingcheng, Ye, Wenjie, Ye, Zhuorui, Yin, Bohong, Yu, Longhui, Yuan, Enming, Yuan, Hongbang, Yuan, Mengjie, Yuan, Siyu, Zhan, Haobing, Zhang, Dehao, Zhang, Hao, Zhang, Wanlu, Zhang, Xiaobin, Zhang, Yadong, Zhang, Yangkun, Zhang, Yichi, Zhang, Yizhi, Zhang, Yongting, Zhang, Yu, Zhang, Yutao, Zhang, Yutong, Zhang, Zheng, Zhao, Haotian, Zhao, Yikai, Zhao, Zijia, Zheng, Huabin, Zheng, Shaojie, Zhong, Longguang, Zhou, Jianren, Zhou, Xinyu, Zhou, Zaida, Zhu, Jinguo, Zhu, Zhen, Zhuang, Weiyu, Zu, Xinxing
Formato:	Preprint
Publicado:	2025
Materias:	Machine Learning Artificial Intelligence Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

_version_	1866917242809090048
author	Kimi Team Bai, Yifan Bao, Yiping Charles, Y. Chen, Cheng Chen, Guanduo Chen, Haiting Chen, Huarong Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, Zhuofu Cui, Jialei Ding, Hao Dong, Mengnan Du, Angang Du, Chenzhuang Du, Dikang Du, Yulun Fan, Yu Feng, Yichen Fu, Kelin Gao, Bofei Gao, Chenxiao Gao, Hongcheng Gao, Peizhong Gao, Tong Ge, Yuyao Geng, Shangyi Gu, Qizheng Gu, Xinran Guan, Longyu Guo, Haiqing Guo, Jianhang Hao, Xiaoru He, Tianhong He, Weiran He, Wenyang He, Yunjia Hong, Chao Hu, Hao Hu, Yangyang Hu, Zhenxing Huang, Weixiao Huang, Zhiqi Huang, Zihao Jiang, Tao Jiang, Zhejun Jin, Xinyi Kang, Yongsheng Lai, Guokun Li, Cheng Li, Fang Li, Haoyang Li, Ming Li, Wentao Li, Yang Li, Yanhao Li, Yiwei Li, Zhaowei Li, Zheming Lin, Hongzhan Lin, Xiaohan Lin, Zongyu Liu, Chengyin Liu, Chenyu Liu, Hongzhang Liu, Jingyuan Liu, Junqi Liu, Liang Liu, Shaowei Liu, T. Y. Liu, Tianwei Liu, Weizhou Liu, Yangyang Liu, Yibo Liu, Yiping Liu, Yue Liu, Zhengying Lu, Enzhe Lu, Haoyu Lu, Lijun Luo, Yashuo Ma, Shengling Ma, Xinyu Ma, Yingwei Mao, Shaoguang Mei, Jie Men, Xin Miao, Yibo Pan, Siyuan Peng, Yebo Qin, Ruoyu Qin, Zeyu Qu, Bowen Shang, Zeyu Shi, Lidong Shi, Shengyuan Song, Feifan Su, Jianlin Su, Zhengyuan Sui, Lin Sun, Xinjie Sung, Flood Tai, Yunpeng Tang, Heyi Tao, Jiawen Teng, Qifeng Tian, Chaoran Wang, Chensi Wang, Dinglu Wang, Feng Wang, Hailong Wang, Haiming Wang, Jianzhou Wang, Jiaxing Wang, Jinhong Wang, Shengjie Wang, Shuyi Wang, Si Wang, Xinyuan Wang, Yao Wang, Yejie Wang, Yiqin Wang, Yuxin Wang, Yuzhi Wang, Zhaoji Wang, Zhengtao Wang, Zhengtao Wang, Zhexu Wei, Chu Wei, Qianqian Wu, Haoning Wu, Wenhao Wu, Xingzhe Wu, Yuxin Xiao, Chenjun Xie, Jin Xie, Xiaotong Xiong, Weimin Xu, Boyu Xu, Jinjing Xu, L. H. Xu, Lin Xu, Suting Xu, Weixin Xu, Xinran Xu, Yangchuan Xu, Ziyao Xu, Jing Xu, Jing Yan, Junjie Yan, Yuzi Yang, Hao Yang, Xiaofei Yang, Yi Yang, Ying Yang, Zhen Yang, Zhilin Yang, Zonghan Yao, Haotian Yao, Xingcheng Ye, Wenjie Ye, Zhuorui Yin, Bohong Yu, Longhui Yuan, Enming Yuan, Hongbang Yuan, Mengjie Yuan, Siyu Zhan, Haobing Zhang, Dehao Zhang, Hao Zhang, Wanlu Zhang, Xiaobin Zhang, Yadong Zhang, Yangkun Zhang, Yichi Zhang, Yizhi Zhang, Yongting Zhang, Yu Zhang, Yutao Zhang, Yutong Zhang, Zheng Zhao, Haotian Zhao, Yikai Zhao, Zijia Zheng, Huabin Zheng, Shaojie Zhong, Longguang Zhou, Jianren Zhou, Xinyu Zhou, Zaida Zhu, Jinguo Zhu, Zhen Zhuang, Weiyu Zu, Xinxing
author_facet	Kimi Team Bai, Yifan Bao, Yiping Charles, Y. Chen, Cheng Chen, Guanduo Chen, Haiting Chen, Huarong Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, Zhuofu Cui, Jialei Ding, Hao Dong, Mengnan Du, Angang Du, Chenzhuang Du, Dikang Du, Yulun Fan, Yu Feng, Yichen Fu, Kelin Gao, Bofei Gao, Chenxiao Gao, Hongcheng Gao, Peizhong Gao, Tong Ge, Yuyao Geng, Shangyi Gu, Qizheng Gu, Xinran Guan, Longyu Guo, Haiqing Guo, Jianhang Hao, Xiaoru He, Tianhong He, Weiran He, Wenyang He, Yunjia Hong, Chao Hu, Hao Hu, Yangyang Hu, Zhenxing Huang, Weixiao Huang, Zhiqi Huang, Zihao Jiang, Tao Jiang, Zhejun Jin, Xinyi Kang, Yongsheng Lai, Guokun Li, Cheng Li, Fang Li, Haoyang Li, Ming Li, Wentao Li, Yang Li, Yanhao Li, Yiwei Li, Zhaowei Li, Zheming Lin, Hongzhan Lin, Xiaohan Lin, Zongyu Liu, Chengyin Liu, Chenyu Liu, Hongzhang Liu, Jingyuan Liu, Junqi Liu, Liang Liu, Shaowei Liu, T. Y. Liu, Tianwei Liu, Weizhou Liu, Yangyang Liu, Yibo Liu, Yiping Liu, Yue Liu, Zhengying Lu, Enzhe Lu, Haoyu Lu, Lijun Luo, Yashuo Ma, Shengling Ma, Xinyu Ma, Yingwei Mao, Shaoguang Mei, Jie Men, Xin Miao, Yibo Pan, Siyuan Peng, Yebo Qin, Ruoyu Qin, Zeyu Qu, Bowen Shang, Zeyu Shi, Lidong Shi, Shengyuan Song, Feifan Su, Jianlin Su, Zhengyuan Sui, Lin Sun, Xinjie Sung, Flood Tai, Yunpeng Tang, Heyi Tao, Jiawen Teng, Qifeng Tian, Chaoran Wang, Chensi Wang, Dinglu Wang, Feng Wang, Hailong Wang, Haiming Wang, Jianzhou Wang, Jiaxing Wang, Jinhong Wang, Shengjie Wang, Shuyi Wang, Si Wang, Xinyuan Wang, Yao Wang, Yejie Wang, Yiqin Wang, Yuxin Wang, Yuzhi Wang, Zhaoji Wang, Zhengtao Wang, Zhengtao Wang, Zhexu Wei, Chu Wei, Qianqian Wu, Haoning Wu, Wenhao Wu, Xingzhe Wu, Yuxin Xiao, Chenjun Xie, Jin Xie, Xiaotong Xiong, Weimin Xu, Boyu Xu, Jinjing Xu, L. H. Xu, Lin Xu, Suting Xu, Weixin Xu, Xinran Xu, Yangchuan Xu, Ziyao Xu, Jing Xu, Jing Yan, Junjie Yan, Yuzi Yang, Hao Yang, Xiaofei Yang, Yi Yang, Ying Yang, Zhen Yang, Zhilin Yang, Zonghan Yao, Haotian Yao, Xingcheng Ye, Wenjie Ye, Zhuorui Yin, Bohong Yu, Longhui Yuan, Enming Yuan, Hongbang Yuan, Mengjie Yuan, Siyu Zhan, Haobing Zhang, Dehao Zhang, Hao Zhang, Wanlu Zhang, Xiaobin Zhang, Yadong Zhang, Yangkun Zhang, Yichi Zhang, Yizhi Zhang, Yongting Zhang, Yu Zhang, Yutao Zhang, Yutong Zhang, Zheng Zhao, Haotian Zhao, Yikai Zhao, Zijia Zheng, Huabin Zheng, Shaojie Zhong, Longguang Zhou, Jianren Zhou, Xinyu Zhou, Zaida Zhu, Jinguo Zhu, Zhen Zhuang, Weiyu Zu, Xinxing
contents	We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.
format	Preprint
id	arxiv_https___arxiv_org_abs_2507_20534
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	Kimi K2: Open Agentic Intelligence Kimi Team Bai, Yifan Bao, Yiping Charles, Y. Chen, Cheng Chen, Guanduo Chen, Haiting Chen, Huarong Chen, Jiahao Chen, Ningxin Chen, Ruijue Chen, Yanru Chen, Yuankun Chen, Yutian Chen, Zhuofu Cui, Jialei Ding, Hao Dong, Mengnan Du, Angang Du, Chenzhuang Du, Dikang Du, Yulun Fan, Yu Feng, Yichen Fu, Kelin Gao, Bofei Gao, Chenxiao Gao, Hongcheng Gao, Peizhong Gao, Tong Ge, Yuyao Geng, Shangyi Gu, Qizheng Gu, Xinran Guan, Longyu Guo, Haiqing Guo, Jianhang Hao, Xiaoru He, Tianhong He, Weiran He, Wenyang He, Yunjia Hong, Chao Hu, Hao Hu, Yangyang Hu, Zhenxing Huang, Weixiao Huang, Zhiqi Huang, Zihao Jiang, Tao Jiang, Zhejun Jin, Xinyi Kang, Yongsheng Lai, Guokun Li, Cheng Li, Fang Li, Haoyang Li, Ming Li, Wentao Li, Yang Li, Yanhao Li, Yiwei Li, Zhaowei Li, Zheming Lin, Hongzhan Lin, Xiaohan Lin, Zongyu Liu, Chengyin Liu, Chenyu Liu, Hongzhang Liu, Jingyuan Liu, Junqi Liu, Liang Liu, Shaowei Liu, T. Y. Liu, Tianwei Liu, Weizhou Liu, Yangyang Liu, Yibo Liu, Yiping Liu, Yue Liu, Zhengying Lu, Enzhe Lu, Haoyu Lu, Lijun Luo, Yashuo Ma, Shengling Ma, Xinyu Ma, Yingwei Mao, Shaoguang Mei, Jie Men, Xin Miao, Yibo Pan, Siyuan Peng, Yebo Qin, Ruoyu Qin, Zeyu Qu, Bowen Shang, Zeyu Shi, Lidong Shi, Shengyuan Song, Feifan Su, Jianlin Su, Zhengyuan Sui, Lin Sun, Xinjie Sung, Flood Tai, Yunpeng Tang, Heyi Tao, Jiawen Teng, Qifeng Tian, Chaoran Wang, Chensi Wang, Dinglu Wang, Feng Wang, Hailong Wang, Haiming Wang, Jianzhou Wang, Jiaxing Wang, Jinhong Wang, Shengjie Wang, Shuyi Wang, Si Wang, Xinyuan Wang, Yao Wang, Yejie Wang, Yiqin Wang, Yuxin Wang, Yuzhi Wang, Zhaoji Wang, Zhengtao Wang, Zhengtao Wang, Zhexu Wei, Chu Wei, Qianqian Wu, Haoning Wu, Wenhao Wu, Xingzhe Wu, Yuxin Xiao, Chenjun Xie, Jin Xie, Xiaotong Xiong, Weimin Xu, Boyu Xu, Jinjing Xu, L. H. Xu, Lin Xu, Suting Xu, Weixin Xu, Xinran Xu, Yangchuan Xu, Ziyao Xu, Jing Xu, Jing Yan, Junjie Yan, Yuzi Yang, Hao Yang, Xiaofei Yang, Yi Yang, Ying Yang, Zhen Yang, Zhilin Yang, Zonghan Yao, Haotian Yao, Xingcheng Ye, Wenjie Ye, Zhuorui Yin, Bohong Yu, Longhui Yuan, Enming Yuan, Hongbang Yuan, Mengjie Yuan, Siyu Zhan, Haobing Zhang, Dehao Zhang, Hao Zhang, Wanlu Zhang, Xiaobin Zhang, Yadong Zhang, Yangkun Zhang, Yichi Zhang, Yizhi Zhang, Yongting Zhang, Yu Zhang, Yutao Zhang, Yutong Zhang, Zheng Zhao, Haotian Zhao, Yikai Zhao, Zijia Zheng, Huabin Zheng, Shaojie Zhong, Longguang Zhou, Jianren Zhou, Xinyu Zhou, Zaida Zhu, Jinguo Zhu, Zhen Zhuang, Weiyu Zu, Xinxing Machine Learning Artificial Intelligence Computation and Language We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.
title	Kimi K2: Open Agentic Intelligence
topic	Machine Learning Artificial Intelligence Computation and Language
url	https://arxiv.org/abs/2507.20534