文｜光锥智能 白鸽 而在大洋彼岸，Claude Sonnet 4.5同期发布。谷歌Gemini 3.0也预计将在10月迎战。 显然，全球大模型进入了一个新的井喷期，各家在大模型产品和技术迭代上，都在明显加速。据不完全 统计，仅9月全球就有15家主流企业/机构发布了开源大模型，数量较8月激增70%。 开源大模型的热潮，也是开发者的盛宴。有开发者都在社交媒体上感慨：一天测了3个模型，还有一个 没测试，这还只是周一...... 这个十一黄金周，AI圈真的很热闹。 10月1日凌晨，OpenAI深夜炸场，正式官宣新一代AI视频模型Sora 2，并直接上线了一个AI版的"抖 音"——一个完全由AI来生成视频的短视频平台。 就在Sora 2发布之前，这一轮的大模型发布其实已经卷起来了。 黄金周前夕，先是国内大模型公司深度求索宣布推出DeepSeek-V3.2-Exp实验版模型，紧随其后智谱AI 正式发布新一代大模型GLM-4.6。 中国大模型也引起了图灵奖得主杨立昆的关注。就在9月30日，蚂蚁集团旗下百灵大模型宣布开源其思 考模型Ring-1T的预览版Ring-1T-preview，参数量达1000B（1万亿）。"令人印象 ...

Core Viewpoint - The competition between open-source and closed-source AI models is intensifying, with open-source models like DeepSeek and Qwen leading China's tech advancement globally. However, concerns about the profitability and economic impact of open-source models persist [1]. Group 1: Open Source vs Closed Source - Open-source software allows community participation beyond organizational boundaries, which is essential for sustainable development and ecological growth [4]. - The distinction between open-source and closed-source languages is significant, with open-source languages like Ruby fostering broader collaboration and innovation [6]. - The coexistence of open-source and closed-source models is expected, with both contributing to competitive software ecosystems [7]. Group 2: Economic Impact of Open Source - Japan's experience with Ruby demonstrates that open-source languages can empower smaller contractors to engage in larger projects, enhancing local economic development [10][11]. - The presence of Ruby's creator in Shimane Prefecture has been pivotal in establishing a local ecosystem that supports larger engineering projects [11]. - The development of a robust open-source community can help retain local talent and stimulate regional economic growth, as seen in Shimane [12]. Group 3: Lessons for China - China can learn from Japan's open-source initiatives to build a new regional economic engine, especially in light of the risks associated with reliance on closed-source AI models [13]. - The importance of open-source algorithms in AI development is emphasized, advocating for a competitive landscape that includes both open-source and closed-source options [13]. - Educational initiatives to promote understanding of open-source principles are crucial for fostering a skilled workforce capable of contributing to open-source projects [16]. Group 4: Challenges and Future of Programming Languages - The rise of AI in programming may lead to a divide between those who understand programming and those who rely solely on AI-generated code, potentially impacting the future of programming languages like Ruby [19][21]. - The need for education in programming remains critical, as reliance on AI could diminish human cognitive skills in understanding IT [21]. - The balance between efficiency gained through AI and the necessity for human understanding of programming concepts is a key consideration for the future [21].

让机器人学会拧开可乐瓶，竟然比让它学会后空翻还要难上十倍？灵巧手成为机器人"最后一公里"的背 后，是困扰行业数十年的性能、成本与可靠性的"不可能三角"。回望四十年技术演进，灵巧手的六大技 术门派都分别是什么？而它们之间又有哪些恩怨故事？在AI时代，大模型又将如何赋能VLA控制和 Sim2Real仿真，加速从实验室到现实应用的跃进。 本期视频我们也将走访硅谷团队现场演示抓螺丝、开可乐、拿iPhone等高难度操作。AI大模型正为灵巧 手注入"智能小脑"，而开源生态可能彻底颠覆传统硬件垄断。未来，机器人能否真正走进家庭？灵巧手 会先在哪一场景普及？这场技术革命，是否将改写人机协作的规则？ ...

来源：鲁网 2025年9月18-20日，华为全联接大会 2025 （HC2025）以 "跃升行业智能化" 为主题，OpenAtom openEuler（简称："openEuler"或"开源欧拉"）作为开源操作系统领域的关键力量深度参与其中。高层演 讲、技术实践、生态成果与合作案例，展现开源操作系统如何打破行业壁垒，从算力基础设施协同到多 产业场景适配，openEuler 正联接起软硬件厂商、企业用户与开发者，用开放协作推动产业资源深度整 合，为各领域融合发展提供新路径。 《创新引领，开源开放，共创智能世界生态新选择》 华为常务董事汪涛在演讲中表示，openEuler历经六年发展，已赋能万千企业应用，服务全球多个国家 和地区，推动构建全球开源新生态，开创了中国开源发展的新范式。同时，openEuler已在下一代操作 系统技术的探索中取得突破：灵衢互联总线技术实现了更高效的资源池化与弹性供给，相关核心能力也 将贡献至openEuler社区。我们相信，依托全球开发者的力量，openEuler将成为用户首选的数智基础设 施操作系统。 华为鲲鹏计算业务总裁 李义 《南网妙算系列高能效服务器技术创新分享》 国家"双碳"战 ...

Group 1 - The profit of industrial enterprises in China improved significantly from a year-on-year decline of 1.7% in the first seven months to a growth of 0.9% in the first eight months of 2025 [2] - The manufacturing sector saw a profit increase of 7.4%, accelerating by 2.6 percentage points compared to the previous month [2] - In August, the profit of industrial enterprises experienced a notable growth of 20.4%, reversing a decline of 1.5% in July [2] Group 2 - During the National Day and Mid-Autumn Festival holiday, small passenger cars will be allowed free passage on national toll roads from October 1 to October 8, 2025 [3] - The free passage applies to vehicles exiting toll roads during the specified time frame, with specific guidelines for ordinary and expressways [3] Group 3 - The scale of the metaverse-related industry in Shanghai has surpassed 300 billion yuan, with over 30 XR terminal manufacturing companies [4] - The Shanghai Future Industry Fund has successfully expanded its scale from 10 billion yuan to 15 billion yuan, focusing on disruptive innovation and early-stage technology investments [4] Group 4 - Li Kaifu emphasized that DeepSeek's core contribution to China's AI development is the promotion of an open-source ecosystem, which is crucial for keeping pace with the U.S. [5] - Tencent has released and open-sourced its new generation multimodal image model, HunyuanImage 3.0, which is the largest and best-performing open-source model in its category [5] Group 5 - Meituan has launched nighttime delivery services using drones in Shenzhen, marking the first integration of drone technology into nighttime instant retail delivery in China [7] - Since its regular operation began in 2021, Meituan's drones have completed over 600,000 delivery tasks, significantly reducing average delivery times [7] Group 6 - The sales revenue of the Xuancheng Pang Donglai Trading Group has exceeded 17 billion yuan in the first nine months of 2025, surpassing its total sales for the entire year of 2024 [8] - The supermarket segment leads in sales, with a notable contribution from the Pang Donglai Times Square store [8] Group 7 - BMW is recalling over 330,000 vehicles due to potential short circuit hazards in starter motors and related components, affecting more than 130,000 cars in Germany alone [11] - The recall involves multiple models produced between September 2015 and September 2021, indicating a potentially broader international impact [11] Group 8 - Eli Lilly's new drug Inluriyo has been approved by the FDA for treating advanced or metastatic breast cancer in specific patient populations [11] - The drug is an oral estrogen receptor antagonist, targeting patients with ER+ and HER2- breast cancer who have progressed after at least one line of endocrine therapy [11]

机器之心报道 编辑：Panda、泽南 AI 行业很多人相信，我们正在或已经进入所谓的「AI 下半场」。在这一轮 AI 的浪潮中，硬件的竞争早已不再是单纯的算力比拼，而是一场围绕软件、开发者与 生态的「护城河」之战。当国产 AI 生态的转型成为科技领域的时代呼声，华为昇腾及其异构计算架构 CANN 正站在了这场变革的聚光灯下。 2025 年 8 月 5 日，华为轮值董事长徐直军宣布， 昇腾硬件使能的 CANN 将全面开源 开 放 ，并承诺在 12 月 30 日 前完成。 从「砖」到「大厦」 这并非一次简单的技术路线调整，而是国产 AI 基础设施在生态战略上一次深刻的自我革命。它标志着昇腾正试图打破过去由硬件厂商自上而下主导的封闭模式， 转而拥抱一种更开放、更依赖社区共建、也更不确定的未来。这不仅是对行业领导者 CUDA 所建立的强大壁垒发起的挑战，更是对自身的一次考验：从「可用」 到「首选」的道路，关键不再仅仅是技术的迭代，更在于 能否真正构建起一个让开发者愿意参与进来的繁荣生态 。 为了探寻 CANN 开源的真实意义，以及它将为整个生态带来怎样的机遇与挑战，我们与三位身处不同位置的核心开发者与观察者进行了深度 ...

Core Insights - The article emphasizes the importance of viewing companies as living entities within an ecosystem, particularly in the context of the accelerating artificial intelligence (AI) era, which necessitates industry collaboration and sustainable development to unlock AI's full potential [1][3] - Huawei is committing to invest 15 billion RMB annually over the next five years to develop its ecosystem, providing 1.5 million PetaFLOPS of computing power to global open-source communities and mobilizing 15,000 people for ecosystem platform development [1][2] Huawei's Ecosystem Strategy - Huawei's ecosystem strategy is evolving from a focus on connectivity to a deeper engagement with terminal and computing industries, expanding its service offerings to developers [2][4] - The company has established a significant presence in the open-source community with projects like OpenHarmony, openEuler, and openGauss, and is actively collaborating with global universities to cultivate future developer talent [2][4][10] Technological Foundations - Key components of Huawei's ecosystem include the HarmonyOS, Kunpeng computing platform, Ascend AI platform, and Huawei Cloud, which together form a comprehensive technological foundation from terminal to cloud [4][5] - The HarmonyOS community has grown to over 9,200 contributors, with more than 1300 hardware products passing compatibility tests, demonstrating significant growth in application ecosystems [5][6] AI and Cloud Services - Huawei has launched the "Tian Gong Plan," investing 1 billion RMB to support AI ecosystem innovation, aiming to incubate over 10,000 AI-native services and 5,000 intelligent agents [5][6] - Huawei Cloud is positioned as a key bridge connecting developers with computing resources, with a developer base that has grown to 8.5 million, emphasizing its role as an incubator for developer ecosystems [6][7] Open Source and Talent Development - The company is committed to open-source principles and talent development, launching initiatives like the "Basic Software Thousand Schools Action" to train 1 million new software talents globally [10][11] - Huawei's open-source projects not only lower development barriers but also foster a collaborative environment for global developers to innovate [10][11] Global Developer Community - Huawei is building a "global developer community" that transcends traditional business models, serving as a platform for technology exchange and the incubation of new applications and standards [12] - The company's strategy aims to position it as a platform enterprise in the global intelligent industry, leveraging open-source and talent development to prepare for the next wave of AI innovation [12]

Core Insights - Huawei unexpectedly revealed its future chip roadmap during the Huawei Connect 2025 event, showcasing several new chips including the Ascend 950, 960, and 970 series for AI computing, as well as the Kunpeng 950 and 960 processors for general computing [1][3][10] Group 1: Chip Developments - The Ascend 950 series chips will support low-precision data formats and achieve computing power of 1P and 2P, enhancing training efficiency and inference throughput [3][10] - The Ascend 960 is planned to double the performance of the Ascend 950 and will support Huawei's self-developed HiF4 data format, set to launch in Q4 2027 [7] - The Ascend 970 will further enhance specifications compared to the Ascend 960, with plans for release in Q4 2028 [7] Group 2: Supernode and Cluster Innovations - Huawei introduced the Atlas 950 supernode, which will consist of 8192 Ascend 950DT chips, achieving FP8 computing power of 8E FLOPS and FP4 computing power of 16E FLOPS, set to launch in Q4 2026 [11][13] - The Atlas 960 supernode, planned for Q4 2027, will be based on 15488 Ascend 960 chips, with FP8 computing power reaching 30E FLOPS and FP4 computing power reaching 60E FLOPS [13] - The Atlas 950 SuperCluster will consist of 64 Atlas 950 supernodes, achieving FP8 computing power of 524 EFLOPS, making it the world's strongest computing cluster [18] Group 3: Software and Ecosystem Development - Huawei aims to develop a robust software ecosystem to complement its hardware, with the CANN deep learning framework and MindSpore framework serving as alternatives to NVIDIA's CUDA [21][22] - The company plans to open-source its CANN compiler and virtual instruction set interface by the end of 2025, along with the Mind series application tools [22][24] - Huawei's strategy emphasizes hardware evolution through existing chip technology while fostering an open-source ecosystem to address challenges posed by U.S. sanctions [24]

Core Insights - The article discusses the "AI Application Industry Monthly Report" for August 2025, highlighting the growth and challenges within the AI application landscape, driven by government initiatives and user engagement metrics [3][4][7]. Group 1: Industry Overview - The "Opinions on Deepening the Implementation of 'Artificial Intelligence +'" was released by the State Council, outlining milestones for 2027, 2030, and 2035, focusing on enhancing foundational model capabilities and data supply innovation [4][7]. - As of August 2025, the user base for native apps in the internet and AI technology sector reached 277 million, while In-App AI users totaled 622 million, indicating a combined AI application user base of 645 million [4][22]. Group 2: User Engagement and Growth Metrics - The top 50 AI applications by monthly active users show that In-App AI applications dominate with 29 entries, followed by PC web applications (9), mobile AI assistants (6), native apps (5), and PC client applications (1) [4][28]. - The leading AI search engine and comprehensive assistant applications have user bases of 672 million and 651 million, respectively, with AI information retrieval emerging as a significant growth area, showing a 39% month-over-month increase [4][26][38]. Group 3: Competitive Landscape - The competitive landscape reveals that while major applications continue to grow, long-tail applications (under 100,000 users) are struggling, with an average month-over-month decline of 6.7% [5][37]. - Mobile AI assistants from manufacturers like OPPO, Xiaomi, and Huawei hold significant market shares, with user engagement rates varying, where Xiaomi's assistant shows a 37.7% active user rate due to its integration with smart home devices [5][58]. Group 4: Application Categories and Challenges - The report categorizes AI applications into five main types, each facing unique challenges: native apps are stable but long-tail applications struggle; In-App AI applications show a clear divide between top and bottom performers; and PC applications are experiencing traffic declines [5][35][46]. - The AI search sector is characterized by a "function easy to popularize, habit hard to cultivate" dilemma, with only three of the top ten search applications achieving over 10% usage [67][69]. Group 5: Future Directions - The future focus for mobile AI assistants is shifting towards enhancing user experience and retention, with a growing emphasis on intelligent data management and personalized services [57][60]. - The integration of AI search with service offerings is becoming a consensus, with the need to stimulate user search behavior being critical for commercial success [73][75].

Group 1 - China's contribution to the global open-source ecosystem for large models has reached 18.7%, ranking second after the US at 37.4%, with both countries accounting for over 55% of the total contributions [1] - The report highlights the dominance of China and the US in the AI open-source field, which may boost investor confidence in domestic AI companies and attract capital inflow [1] - The trend indicates that the open-source ecosystem is becoming a core aspect of technological competition, while also posing risks such as technological blockades and collaboration barriers [1] Group 2 - NVIDIA is reportedly scaling back its cloud computing business, specifically reducing efforts to attract enterprises to its DGX Cloud service [2] - This move reflects the intense competition in the cloud computing market and indicates NVIDIA's refocus on its core strengths in AI chip design and software ecosystem rather than direct cloud service competition [2] - For investors, this could be a positive signal as it suggests the company is concentrating on higher-margin hardware business, while the industry may need to reassess the competitive landscape of AI infrastructure [2] Group 3 - Dasin Intelligent has signed a contract worth 23.869 million yuan for the Xiaomi Wuhan Phase II smart project, which represents 0.75% of the company's audited revenue for 2024 [3] - Although the contract amount is relatively small, it is expected to positively influence investor sentiment and may lead to brand premium for the company [3] - The trend in the industry shows that smart manufacturing and IoT applications are becoming key drivers for the intelligent upgrade of buildings, with such projects likely to help the company secure more similar orders [3]