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

作者：DeepSeek-AI, Aixin Liu, Aoxue Mei, Lin, Bangcai, Bing Xue, Bingxuan Wang, Bingzheng Xu, Bowen Wu, Bowei Zhang, Chaofan Lin, Chen Dong, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenhao Xu, Chong Ruan, Damai Dai, Daya Guo, Yang Dejian, Deli Chen, Erhang Li, Fangqi Zhou, Fangyun Lin, Dai, Fucong, Guangbo Hao, Guan-Ting Chen, Guowei Li, Hengyun Zhang, Hanwei Xu, Hao Li, Liang, Haofen, Haoran Wei, Haowei Zhang, Haowen Luo, Haozhe Ji, Honghui Ding, Hongxuan Tang, Huanqi Cao, Huazuo Gao, Hui‐Qi Qu, Hui Zeng, Jialiang Huang, Jiashi Li, Jiaxin Xu, Jiewen Hu, Jingchang Chen, Xiang, Jingting, Jingyang Yuan, Jingyuan Cheng, Jinhua Zhu, Jun Ran, Junguang Jiang, Junjie Qiu, Junlong Li, Junxiao Song, Kai Dong, Kaige Gao, Kang Guan, Kexin Huang, Kexing Zhou, Kezhao Huang, Kuai Yu, Lean Wang, Lecong Zhang, Lei Wang, Liang Zhao, Liangsheng Yin, Lihua Guo, Lingxiao Luo, L. R. Ma, Litong Wang, Liyue Zhang, Miaomiao Di, Mingjun Xu, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Mingxu Zhou, Panpan Huang, Peixin Cong, Peiyi Wang, Qiancheng Wang, Qihao Zhu, Qingyang Li, Qinyu Chen, Qiushi Du, R. Xu, Ruiqi Ge, Ruisong Zhang, Pan, Ruizhe, Runji Wang, Yin, Runqiu, Runxin Xu, Shen, Ruomeng, Ruoyu Zhang, S. H. 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, Zhou Songyang, Tao Ni, Yun Tao, Pei, Tian, Ye Tian, Yue, Tianyuan, Wangding Zeng, Wenzhao Liu, Wenfeng Liang, Wenjie Pang, Wenjing Luo, Wen‐Jun Gao, Wentao Zhang, Xiqi Gao, Xiangwen Wang, XiaoGang Bi, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaokang Zhang, Xiaotao Nie, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xingkai Yu, Xingyou Li, Xinyu Yang, Xinyuan Li, Xu Chen, X.-J Su, Xuehai Pan, Xuheng Lin, Xuwei Fu, Yufei Wang, Yang Zhang, Yaobo Xu, Yanru Ma, Yao Li, Yao Li, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yi Xian Qian, Yi Yu, Yichao Zhang, Yifan Ding, Shi, Yifan, Yiliang Xiong, Yinghe He, Ying Zhou, Yinmin Zhong, Yishi Piao, Yisong Wang, Yixiao Chen, Yixuan Tan, Yixuan Wei, Yiyang Ma, Yiyuan Liu, Yang, Yonglun, Yongqiang Guo, Yongtong Wu, Yu Wu, Yuan Cheng, Yangjie Ou, Yuanfan Xu, Yuduan Wang, Yue Gong, Yuhan Wu, Yuheng Zou, Y. Li, Xiong, Yunfan, Yuxiang Luo, Yuxiang You, Yuxuan Liu, Yuyang Zhou, Zhihua Wu, Z. Z. Ren, Zehua Zhao, Zehui Ren, Zhangli Sha, Zhe Fu, Z. L. Xu, Zhenda Xie, Zhengyan 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, Zilin Li, Zipeng Zhang, Ziwei Xie, Gao, Ziyi, Pan, Zizheng, Zhonghua Yao, Bei Feng, Hui Li, Jiali Cai, Jiaqi Ni, Lei Xu, Li Meng, Ning Tian, R. J. Chen, Rong Jin, S. S. Li, Shuang Zhou, Tianyu Sun, X. Q. Li, Jin, Xiangyue, Shen, Xiaojin, Xiaosha Chen, Xinnan Song, Xinyi Zhou, Yuxin Zhu, Yanping Huang, Yaohui Li, Yi Zheng, Yi Zhu, Y. Ma, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, Dongjie Ji, Jian Liang, Jianzhong Guo, Jin Chen, L. Xia, Miaojun Wang, Mingming Li, Peng Zhang, Ruyi Chen, S. S. Sun, Shaoqing Wu, Shengfeng Ye, Ting Wang, Wenlian Xiao, Wei Guang An, X.G. Wang, Xiao-Wen Sun, Xiaoxiang Wang, Ying Tang, Yukun Zha, Zekai Zhang, Zhenyu Ju, Zhen Zhang, Zihua Qu · 发表于：arXiv (Cornell University) · 年份：2025 · DOI：10.48550/arxiv.2512.02556 · 被引用次数：8 · 研究领域：Topic Modeling、Machine Learning in Materials Science、Artificial Intelligence in Healthcare and Education

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.