DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models

作者：DeepSeek-AI, A. Liu, Aoxue Mei, B. Lin, Bing Xue, Bing-Li Wang, Bin Xu, Bochao Wu, Bowei Zhang, Chaofan Lin, Chen Dong, Chengda Lu, Chenggang Zhao, C. Deng, Chenhao Xu, C. Ruan, Damai Dai, Daya Guo, Dejian Yang, Deli Chen, Erhang Li, Fangqi Zhou, Fangyun Lin, Fucong Dai, Guangbo Hao, Guanting Chen, Guowei Li, H. Zhang, Hanwei Xu, Hao Li, Hao Liang, Haoran Wei, Haowei Zhang, Hao-sheng Luo, Haozhe Ji, Honghui Ding, Hongxuan Tang, Huan Cao, Huazuo Gao, Huixian Qu, Hui Zeng, Jialiang Huang, Jiashi Li, Jiaxin Xu, Jiewen Hu, JingChang Chen, Ji Xiang, Jingyang Yuan, Jing Cheng, Jinhua Zhu, Jun Ran, Junguang Jiang, Junjie Qiu, Junlong Li, Jun-Mei Song, Kai Dong, Kaige Gao, Kang Guan, Kexin Huang, Kexin Zhou, Ke-wei Huang, Kuai Yu, Lean Wang, Lecong Zhang, Lei Wang, Liang Zhao, Liangsheng Yin, Lihua Guo, Ling-Li Luo, Lin Ma, Litong Wang, Liyue Zhang, M. Di, M. Y. Xu, Mingchuan Zhang, Minghua Zhang, M. Tang, Mingxu Zhou, P. Huang, Peixin Cong, Peiyi Wang, Qiancheng Wang, Qihao Zhu, Qingyang Li, Qinyu Chen, Qiushi Du, Ruiling Xu, Ruiqi Ge, Ruisong Zhang, Ruizhe Pan, Runji Wang, Runqiu Yin, Runxin Xu, Ru Shen, Ruoyu Zhang, S. Liu, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shaofei Cai, Shaoyuan Chen, Shengding Hu, Shengyu Liu, Shiqiang Hu, Shirong Ma, Shiyu Wang, Shuiping Yu, Shunfeng Zhou, Shuting Pan, Songyang Zhou, Tao Ni, Tao Yun, Tian Pei, Tian Ye, Tianyuan Yue, Wangding Zeng, Wen Liu, W. Liang, Wenjie Pang, Wenjing Luo, Wenjun Gao, Wentao Zhang, Xi Gao, Xiangwen Wang, Xiaoling Bi, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaokang Zhang, X. Nie, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xingkai Yu, Xingyou Li, Xinyu Yang, Xinyuan Li, Xu Chen, Xuecheng Su, Xuehai Pan, Xuheng Lin, Xu Fu, Y. Q. Wang, Yang Zhang, Yanhong Xu, Yanru Ma, Yao Li, Yao Zhao, Yaofeng Sun, Yaohui Wang, Y. Qian, Yingpu Yu, Yichao Zhang, Yifan Ding, Yifan Shi, Yi Xiong, Ying He, Ying Zhou, Yinmin Zhong, Y. Piao, Yisong Wang, Yixiao Chen, Yixuan Tan, Yixuan Wei, Yiyang Ma, Yiyuan Liu, Yong Yang, Yongqiang Guo, Yongtong Wu, Yu Wu, Yuan Cheng, Y. Ou, Yuanfang Xu, Yuduan Wang, Yue Gong, Yuhan Wu, Yu-Hui Zou, Yukun Li, Yunfan Xiong, Yu-Wei Luo, Yu-mei You, Yuxuan Liu, Yuyang Zhou, Z. F. Wu, Z. Ren, Zehua Zhao, Z. Ren, Zhangli Sha, Zhe Fu, Zhean Xu, Zhenda Xie, Zhen-guo Zhang, Zhewen Hao, Zhibin Gou, Zhicheng Ma, Zhigang Yan, Zhihong Shao, Zhixian Huang, Zhiyu Wu, Zhuoshu Li, Zhuping Zhang, Zian Xu, Zihao Wang, Zihui Gu, Zijia Zhu, Ziling Li, Zipeng Zhang, Ziwei Xie, Ziyi Gao, Zizheng Pan, Z. Yao, B. Feng, Hui Li, J. Cai, J. Ni, Lei Xu, Meng Li, Ning Tian, R. J. Chen, R. Jin, S. Li, Shuang Zhou, T. Sun, X. Q. Li, Xiangyu Jin, Xiaojin Shen, Xiaosha Chen, Xinnan Song, Xinyi Zhou, Y. X. Zhu, Yanping Huang, Yao Li, Yi Zheng, Yuchen Zhu, Yunxiang Ma, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, Dong-Li Ji, Jian Liang, Jianzhong Guo, Jin Chen, Leyi Xia, Miaojun Wang, Mingming Li, Peng Zhang, Ruyi Chen, Shangmian Sun, Shao-Kang Wu, Sheng-Ying Ye, T.Wang, W. Xiao, Wei An, Xianzu Wang, Xiaowen Sun, Xiaoxiang Wang, Ying Tang, Y. Zha, Ze-Na Zhang, Zhenghua Ju, Zhen Zhang, Z. Qu · 发表于：arXiv.org · 年份：2025 · DOI：10.48550/arXiv.2512.02556 · 被引用次数：678 · 研究领域：Computer Science

We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. The key technical breakthroughs of DeepSeek-V3.2 are as follows: (1) DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance in long-context scenarios. (2) Scalable Reinforcement Learning Framework: By implementing a robust reinforcement learning protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par with Gemini-3.0-Pro, achieving gold-medal performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). (3) Large-Scale Agentic Task Synthesis Pipeline: To integrate reasoning into tool-use scenarios, we developed a novel synthesis pipeline that systematically generates training data at scale. This methodology facilitates scalable agentic post-training, yielding substantial improvements in generalization and instruction-following robustness within complex, interactive environments.