Step 3.5 Flash:拥有 11B 激活参数的开放前沿级智能模型

发表
taesiritaesiri 提交
作者: Ailin Huang, Ang Li, Aobo Kong, Bin Wang, Binxing Jiao, Bo Dong, Bojun Wang, Boyu Chen, Brian Li, Buyun Ma, Chang Su, Changxin Miao, Changyi Wan, Chao Lou, Chen Hu, Chen Xu, Chenfeng Yu, Chengting Feng, Chengyuan Yao, Chunrui Han, Dan Ma, Dapeng Shi, Daxin Jiang, Dehua Ma, Deshan Sun, Di Qi, Enle Liu, Fajie Zhang, Fanqi Wan, Guanzhe Huang, Gulin Yan, Guoliang Cao, Guopeng Li, Han Cheng, Hangyu Guo, hanshanzhangHanshan Zhang, Hao Nie, Haonan Jia, Haoran Lv, Hebin Zhou, Hekun Lv, Heng Wang, Heung-Yeung Shum, Hongbo Huang, Hongbo Peng, Hongyu Zhou, Hongyuan Wang, Houyong Chen, Huangxi Zhu, Huimin Wu, Huiyong Guo, Jia Wang, Jian Zhou, Jianjian Sun, Jiaoren Wu, Jiaran Zhang, Jiashu Lv, Jiashuo Liu, Jiayi Fu, Jiayu Liu, Jie Cheng, Jie Luo, Jie Yang, Jie Zhou, Jieyi Hou, Jing Bai, Jingcheng HuJingcheng Hu, Jingjing Xie, Jingwei Wu, Jingyang Zhang, Jishi Zhou, Junfeng Liu, Junzhe Lin, Ka Man Lo, Kai Liang, Kaibo Liu, Kaijun Tan, Kaiwen YanKaiwen Yan, Kaixiang Li, Kang An, LinkanghengKangheng Lin, Lei Yang, Liang Lv, Liang Zhao, Liangyu Chen, Lieyu Shi, Liguo Tan, Lin Lin, Lina Chen, Luck Ma, Mengqiang Ren, Michael Li, Ming Li, Mingliang Li, Mingming Zhang, Mingrui Chen, Mitt Huang, Na Wang, Peng Liu, Qi Han, Qian Zhao, Qinglin He, Qinxin Du, Qiuping Wu, Quan Sun, Rongqiu Yang, Ruihang Miao, Ruixin Han, Ruosi Wan, Ruyan Guo, Shan Wang, Shaoliang Pang, Shaowen Yang, Shengjie Fan, Shijie Shang, Shiliang Yang, Shiwei Li, Shuangshuang Tian, Siqi Liu, Siye WuSiye Wu, Siyu Chen, Song Yuan, Tiancheng Cao, Tianchi Yue, Tianhao Cheng, Tianning Li, Tingdan Luo, Wang You, Wei Ji, Wei Yuan, Wei Zhang, Weibo Wu, xieweihaoWeihao Xie, Wen Sun, Wenjin Deng, wenWenzhen Zheng, Wuxun Xie, Xiangfeng Wang, Xiangwen Kong, Xiangyu Liu, Xiangyu Zhang, Xiaobo Yang, Xiaojia Liu, Xiaolan Yuan, Xiaoran Jiao, Xiaoxiao Ren, Xiaoyun Zhang, Xin Li, Xin Liu, Xin Wu, Xing Chen, Xingping Yang, Xinran Wang, Xu Zhao, xuan heXuan He, Xuanti Feng, Xuedan Cai, Xuqiang Zhou, Yanbo Yu, Yang Li, Yang Xu, Yanlin Lai, Yanming Xu, Yaoyu Wang, Yeqing Shen, Yibo Zhu, Yichen Lv, Yicheng Cao, Yifeng Gong, Yijing Yang, Yikun Yang, Yin Zhao, Yingxiu Zhao, Yinmin Zhang, Yitong Zhang, Yixuan Zhang, Yiyang Chen, Yongchi Zhao, Yongshen Long, Yongyao Wang, Yousong Guan, Yu Zhou, Yuang Peng, Yuanhao Ding, Yuantao Fan, Yuanzhen Yang, Yuchu Luo, Yudi Zhao, Yue Peng, Yueqiang Lin, Yufan Lu, Yuling Zhao, Yunzhou Ju, Yurong Zhang, Yusheng Li, Yuxiang Yang, Yuyang Chen, Yuzhu Cai, Zejia Weng, Zetao Hong, Zexi Li, Zhe Xie, Zheng Ge, Zheng Gong, Zheng Zeng, zyZhenyi Lu, Zhewei Huang, Zhichao Chang, Zhiguo Huang, Zhiheng Hu, Zidong Yang, Zili Wang, Ziqi Ren, Zixin Zhang, Zixuan Wang

摘要

AI 生成总结
Step 3.5 Flash 是一款稀疏混合专家(MoE)模型,通过高效的参数利用和优化的注意力机制实现了前沿级的智能体智能,在多个基准测试中展现出强劲性能。
我们推出了 Step 3.5 Flash,这是一款兼具前沿智能体能力与计算效率的稀疏混合专家(MoE)模型。在构建智能体时,我们专注于最关键的要素:敏锐的推理能力以及快速、可靠的执行力。Step 3.5 Flash 将 196B 参数的基础规模与 11B 激活参数相结合,实现了高效推理。它采用了 3:1 交错式滑动窗口/全注意力机制以及多 Token 预测(MTP-3)进行优化,从而降低了多轮智能体交互的延迟和成本。为了达到顶尖的智能水平,我们设计了一个可扩展的强化学习框架,将可验证信号与偏好反馈相结合,并在大规模离线策略(off-policy)训练下保持稳定,实现了在数学、代码和工具使用方面的持续自我提升。Step 3.5 Flash 在智能体、编程和数学任务中表现强劲,在 IMO-AnswerBench 上达到 85.4%,在 LiveCodeBench-v6 (2024.08-2025.05) 上达到 86.4%,在 tau2-Bench 上达到 88.2%,在 BrowseComp(含上下文管理)上达到 69.0%,在 Terminal-Bench 2.0 上达到 51.0%,性能媲美 GPT-5.2 xHigh 和 Gemini 3.0 Pro 等前沿模型。通过重新定义效率边界,Step 3.5 Flash 为在真实工业环境中部署复杂的智能体提供了高密度的基础支撑。
查看 arXiv 页面查看 PDF

评论

Lei YangLei Yang

Step-3.5-Flash 在 MathArena(一个防作弊的数学竞赛基准测试)上排名 #1

image

dengwenjindengwenjin

Step 3.5 Flash 官方模型博客:👉 https://static.stepfun.com/blog/step-3.5-flash/