格隆汇9月30日丨寒武纪(688256.SH)公布定增发行情况报告书，根据投资者申购报价情况，并严格按照认购邀请书中确定 的发行价格、发行对象及获配股份数量的程序和规则，确定本次发行价格为1,195.02元/股。本次发行对应的认购总股数为 3,334,946股，认购总金额为3,985,327,168.92元。本次发行对象最终确定为13家。 | 序 | 发行对象名称 | 获配股数 | 获配金额 | 限售期 | | --- | --- | --- | --- | --- | | 름 | | (股) | (元) | (月) | | 1 | 广发基金管理有限公司 | 1.010.861 | 1,207.999.112.22 | 6 | | 2 | UBS AG | 364,010 | 434,999,230.20 | 6 | | 3 | 新华资产管理股份有限公司 | 351,458 | 419,999,339.16 | 6 | | 4 | 汇添富基金管理股份有限公司 | 296,647 | 354,499,097.94 | 6 | | 5 | 诺德基金管理有限公司 | 233,217 | 278,698,979.34 ...

Core Points - The company announced a private placement of 3.3349 million A-shares at a price of RMB 1195.02 per share, raising a total of RMB 3.985 billion [1] - After deducting issuance costs of RMB 32.1713 million (excluding tax), the net proceeds amount to RMB 3.953 billion [1] - The raised funds will be allocated to projects related to a chip platform for large models, a software platform for large models, and to supplement working capital [1]

Investment Rating - The report maintains a "Recommendation" rating for the computer industry, expecting the industry index to rise more than 5% over the next 3-6 months compared to the benchmark index [15]. Core Insights - The release of DeepSeek V3.2-Exp model is expected to drive the development of the domestic AI ecosystem, showcasing the synergy between domestic chips and large models, which leads to significant improvements in computational efficiency and cost reduction [5]. - The introduction of the DeepSeek Sparse Attention mechanism allows for a substantial enhancement in long text processing efficiency, overcoming traditional computational complexity limitations [5]. - The cost of AI applications is anticipated to decrease significantly, with the DeepSeek API prices reduced by over 50%, making advanced AI models more accessible to developers and businesses [5]. - Investment focus is recommended on domestic chip manufacturers like Cambricon and Hygon, as well as companies within the Huawei supply chain and various AI application firms [5]. Industry Basic Data - The computer industry comprises 337 listed companies with a total market capitalization of approximately 61,619.92 billion and a circulating market capitalization of about 55,710.57 billion [2]. Relative Index Performance - The absolute performance of the computer industry over the past 1 month, 6 months, and 12 months is -1.6%, 24.3%, and 72.1% respectively, while the relative performance is -4.3%, 6.3%, and 47.3% [3].

Core Viewpoint - DeepSeek has officially released the V3.2-Exp model, introducing the DeepSeek Sparse Attention (DSA) mechanism, which optimizes training and inference efficiency for long texts, significantly reducing service costs by over 50% for the DeepSeek API [1][5]. Group 1: Model Development - The V3.2-Exp model builds on the V3.1-Terminus version and incorporates the DSA mechanism, which is a sparse attention approach that reduces computational complexity when processing long texts [1][4]. - DSA allows for adaptive selection of key attention heads and local context windows, improving efficiency and lowering costs compared to traditional dense attention mechanisms [3][4]. Group 2: Cost and Accessibility - The introduction of the new model has led to a significant reduction in the cost of accessing the DeepSeek API, with prices dropping by more than 50% [5]. - DeepSeek has temporarily retained additional API access for the previous V3.1-Terminus model until October 15, allowing users to conduct comparative testing [2]. Group 3: Open Source and Community Engagement - DeepSeek has fully open-sourced the V3.2-Exp model on platforms like HuggingFace and ModelScope, along with related research papers [2]. - The company has also open-sourced the TileLang version of the operators, which has garnered significant attention in the industry [1][6]. Group 4: Hardware Compatibility - Following the release of V3.2-Exp, major domestic hardware companies like Huawei, Cambricon, and Haiguang have announced compatibility with the new model, indicating a collaborative development within the domestic AI ecosystem [6][10]. - TileLang, a programming language developed for simplifying GPU operator development, has been recommended for use in research experiments, enhancing the efficiency of AI operator development [7][10].

寒武纪9月30日公告，本次向特定对象发行股票的数量为333.49万股，发行价格为1195.02元/股。发行募 集资金总额为39.85亿元，募集资金净额为39.53亿元。本次发行对象为13家，UBS AG获配43.5亿元，广 发基金管理有限公司获配12.08亿元，诺德基金管理有限公司获配2.79亿元。 ...

中科寒武纪科技股份有限公司 2025 年度向特定对象发行 A 股股票 全体董事签名： 陈天石 刘新宇 金晓光 发行情况报告书 保荐人（主承销商） 广东省深圳市福田区中心三路 8 号卓越时代广场（二期）北座 二〇二五年九月 发行人全体董事、监事及高级管理人员声明 本公司全体董事、监事及高级管理人员承诺本发行情况报告书不存在虚假记 载、误导性陈述或重大遗漏，并对其真实性、准确性、完整性承担相应的法律责 任。 刘少礼 王 在 叶淏尹 胡燏翀 吕红兵 王秀丽 中科寒武纪科技股份有限公司 年 月 日 1 2 发行人全体董事、监事及高级管理人员声明 本公司全体董事、监事及高级管理人员承诺本发行情况报告书不存在虚假记 载、误导性陈述或重大遗漏，并对其真实性、准确性、完整性承担相应的法律责 任。 全体监事签名： 王敦纯 李 一 胡 帅 中科寒武纪科技股份有限公司 年 月 日 刘 毅 王 在 叶淏尹 3 发行人全体董事、监事及高级管理人员声明 本公司全体董事、监事及高级管理人员承诺本发行情况报告书不存在虚假记 载、误导性陈述或重大遗漏，并对其真实性、准确性、完整性承担相应的法律责 任。 全体高级管理人员签名： 陈天石 陈 帅 ...

证券代码：688256 证券简称：寒武纪 公告编号：2025-057 本公司董事会及全体董事保证本公告内容不存在任何虚假记载、误导性陈述 或者重大遗漏，并对其内容的真实性、准确性和完整性依法承担法律责任。 中科寒武纪科技股份有限公司董事会 2025 年 10 月 1 日 关于 2025 年度向特定对象发行股票发行情况的提示性 公告 中科寒武纪科技股份有限公司（以下简称"公司"）2025 年度向特定对象发行 A 股股票（以下简称"本次发行"）发行承销总结相关文件已经上海证券交易所备 案通过，公司将尽快办理本次发行新增股份的登记托管手续。 本次发行的具体情况详见公司同日在上海证券交易所网站(www.sse.com.cn) 披露的《中科寒武纪科技股份有限公司 2025 年度向特定对象发行 A 股股票发行情 况报告书》等相关文件。 特此公告。 中科寒武纪科技股份有限公司 ...

您可使用手机"扫一扫"或进入"注册会计师行业结 目 录 | 、验资报告 …………………………………………………………………………………… 第 1―2 页 | | --- | | 二、附件 · | | --- | | (一)注册资本及实收股本变更前后对照表 …………………………………………………第３页 | | (二)验资事项说明 ……………………………………………………………… 第 4—6 页 | | (三)银行询证函和进账单复印件………………………………………… 第 7—13 页 | | (四)本所营业执照复印件 ································································· 第 14 页 | | (五)本所执业证书复印件 ·································································· 第 15 页 | | (六) 本所《从事证券服务业备案完备证明材料》复印件 …… 第 16 页 | | (七) 签字注册会计师执业证书复印件……………………………第 17—18 页 | 验 ...

北京市中伦律师事务所 关于中科寒武纪科技股份有限公司 2025 年度向特定对象发行 A 股股票 发行过程及认购对象合规性的 法律意见书 二〇二五年九月 北京 • 上海 • 深圳 • 广州 • 武汉 • 成都 • 重庆 • 青岛 • 杭州 • 南京 • 海口 • 东京 • 香港 • 伦敦 • 纽约 • 洛杉矶 • 旧金山 • 阿拉木图 Beijing • Shanghai • Shenzhen • Guangzhou • Wuhan • Chengdu • Chongqing • Qingdao • Hangzhou • Nanjing • Haikou • Tokyo • Hong Kong • London • New York • Los Angeles • San Francisco • Almaty | 一、本次发行的授权和批准 | 5 | | --- | --- | | （一） 发行人的内部授权和批准 | 5 | | （二） 上交所审核及中国证监会注册 | 6 | | 二、本次发行过程的合规性 | 6 | | （一）发出认购邀请 6 | | | （二）投资者申购报价 7 | | | （三）定价和配售 ...

Core Insights - The article discusses the launch of the new AI model GLM-4.6 by Zhipu, highlighting significant advancements in various capabilities and its compatibility with domestic chips [1][2][3] Group 1: Model Enhancements - GLM-4.6 features improvements in advanced coding capabilities, aligning its code performance with Claude Sonnet 4 in public benchmarks and real programming tasks [1] - The context length has been increased from 128K to 200K, allowing for better handling of longer code and intelligent agent tasks [1] - The model's reasoning ability has been enhanced, now supporting tool invocation during inference [2] - Search capabilities have been improved, enabling better tool invocation and search intelligence within the intelligent agent framework [2] - Writing capabilities have been upgraded, making the model's style, readability, and role-playing scenarios more aligned with human preferences [3] Group 2: Commercialization and Market Activity - GLM-4.6 has been launched on the Zhipu MaaS platform and will be open-sourced on Hugging Face and ModelScope under the MIT license [3] - Following the release of the previous model GLM-4.5, Zhipu's MaaS platform API commercialization saw over a tenfold increase [3] - The AI sector is experiencing heightened activity, with stocks of companies like Haitai Ruisheng, Dou Shen Education, and Fabben Information seeing significant gains [8] - Recent collaborations include Kevin Education's joint venture with Zhipu to develop AI educational products, with a registered capital of 50 million yuan [8] - Tuo Si Da announced its humanoid robot "Xiao Tuo," developed in collaboration with Zhipu, capable of autonomous reasoning and decision-making for complex tasks [8] - UCloud has signed a cooperation agreement with Zhipu to enhance model deployment, cloud service optimization, and industry application [8]