（原标题：新一代中国操作系统发布；星舰第十次试飞成功丨新鲜早科技） 21世纪经济报道新质生产力研究院综合报道 早上好，新的一天又开始了。在过去的24小时内，科技行业发生了哪些有意思的事情？来跟21tech一起看看吧。 【巨头风向标】 谷歌正式发布图像生成模型Gemini 2.5 Flash Image 当地时间8月26日，谷歌推出其最先进的图像生成与编辑模型Gemini 2.5 Flash Image，代号"纳米香蕉"（nano banana）。该模型当前在LMArena基 准测试中位列AI图像编辑模型榜首，具备角色一致性保持、自然语言精准修图、多图融合能力，并利用Gemini世界知识提升智能表现。目前用户 可通过Gemini App、API等方式访问，其API定价为每百万输出token30美元。按官方说法，生成单张图片约消耗1290个输出token，折算成本约 0.039美元。 SpaceX星舰第十次试飞成功 连续推迟两天后，SpaceX新一代重型运载火箭"星舰"于美国中部时间8月26日18时30分从得克萨斯州发射升空，实施第十次试飞。据悉，本次任务 重点目标包括让飞船部署模拟卫星、在太空中进行发动机重新点 ...

传媒行业 行业研究 | 动态跟踪 重磅 AI+政策落地，AI 产业链突破将至 核心观点 投资建议与投资标的 1）AI 应用，重点看两个核心赛道：AI+视频图片的快手-W(01024，买入)、美图公司 (01357，未评级)；AI+招聘的 BOSS 直聘-W(02076，买入)、同道猎聘(06100，未评 级)、北森控股(09669，未评级)。 2） 大 厂 （META 逻 辑 演 绎+基 模+算 力 ）， 阿里巴巴-W(09988，买入)、腾讯控股 (00700，买入)、快手-W(01024，买入)、百度集团-SW(09888，买入)、字节跳动（未上 市）。 3）产业链数据及版权，人民网(603000，未评级)、浙数文化(600633，未评级)、视觉中 国(000681，未评级)、中文在线(300364，未评级)。 风险提示：AI 技术迭代不及预期、AI 应用落地不及预期、AI 商业化变现不及预期等 有关分析师的申明，见本报告最后部分。其他重要信息披露见分析师申明之后部分，或请与您的投资代表联系。并请阅读本证券研究报告最后一页的免责申明。 国家/地区 中国 行业 传媒行业 报告发布日期 2025 年 08 月 ...

本文核心观点：2025年中国人工智能代理行业将呈现"技术突破+场景深耕"双轮驱动格局，市场规模预计以 72.7%的年复合增长率爆发式增长，并在技术架构、产品形态、行业应用和政策治理四个维度形成差异化竞 争路径。 中国人工智能代理发展趋势分析 从"模型垄断"到"普惠化Agent"的进化之路 中国人工智能代理行业的技术发展呈现"基础模型突破-架构创新-效率优化"的多层次演进路径，各层次间相 互驱动、协同发展，共同塑造行业技术生态。在基础模型突破层面，大模型能力的跃升与开源化趋势成为核 心驱动力, 推动AI技术从少数企业垄断向行业普惠化发展。 多模态融合技术进一步扩展模型边界推动AI代理 从单一文本交互向多感官感知进化。基础模型的认知能力升维还体现在从被动工具向主动Agent的转型，形 成"感知-决策-执行"闭环。 在架构创新层面，混合架构与协作机制成为技术突破的关键方向。MoA(Mixture-of-Agents)混合架构已成为 行业标准，整合通用模型、垂直场景专业模型、工具链平台及数据飞轮四大组件，其中场景化专业模型通过 蒸馏、微调从基座模型获得专业能力，规模控制在百亿参数以内，特定任务准确率比通用模型高15% ...

Core Insights - The report evaluates the technical capabilities of four major AI model development platforms: Alibaba Cloud's Bailian, Tencent Cloud's Intelligent Agent Development Platform, Kouzi, and Baidu Intelligent Cloud Qianfan, focusing on RAG capabilities, workflow capabilities, and agent capabilities [1][7][8]. RAG Capability Testing - RAG capability testing assesses knowledge enhancement mechanisms, including multi-modal knowledge processing, task complexity adaptation, and interaction mechanism completeness [7][8]. - In text question answering, all platforms demonstrated high accuracy, with over 80% accuracy in multi-document responses, although some platforms showed stability issues during API calls [20][21]. - Baidu Intelligent Cloud Qianfan exhibited stable performance in complex query scenarios for structured data, while Tencent Cloud achieved 100% refusal for out-of-knowledge-base questions [21][23]. - The platforms showed differences in handling refusal and clarification, with Tencent Cloud providing 100% refusals for non-knowledge-base questions [21][22]. Workflow Capability Testing - Workflow capability testing focuses on dynamic parameter extraction, exception rollback, intent recognition, and fault tolerance [35][36]. - The end-to-end accuracy for workflow processes ranged from 61.5% to 93.3%, with Tencent Cloud leading in intent recognition accuracy at 100% [36][37]. - The platforms demonstrated basic usability in workflow systems, but there is room for improvement in complex information processing [38][39]. Agent Capability Testing - Agent capability testing evaluates the ability to call tools, focusing on intent understanding, operational coordination, feedback effectiveness, and mechanism completeness [44][45]. - All platforms achieved high single-tool call completion rates (83%-92%), but multi-tool collaboration and prompt calling showed potential for improvement [48][50]. - Tencent Cloud's Intelligent Agent Development Platform excelled in tool call success rates due to its robust ecosystem and process optimization [49][50].

Core Viewpoint - The article emphasizes that artificial intelligence (AI) has transitioned from being an abstract concept to a tangible force that is increasingly penetrating everyday life, showcasing a significant shift towards practical applications and user engagement in the AI sector [21][23][70]. Group 1: Event Overview - The World Artificial Intelligence Conference (WAIC) attracted a record number of attendees, with an estimated 350,000 visitors, including 156 purchasing groups from over ten countries [4][12]. - The event implemented strict entry protocols, requiring advance ticket purchases and identity verification, reflecting the high demand and interest in AI technologies [5][6]. Group 2: AI Applications and Trends - The focus of AI development has shifted from merely enhancing computational power to applying AI in real-world scenarios, with companies competing to convert technology into productivity [10][21]. - The exhibition featured a wide array of robots capable of performing various tasks, indicating a growing interest in practical applications of robotics in everyday life [10][11][12]. Group 3: User Engagement and Market Response - Ordinary visitors showed significant interest in AI technologies, with many expressing a desire for more affordable robotic solutions after witnessing their capabilities [8][11]. - The popularity of AI products, such as smart glasses, highlights a trend where consumers are eager to engage with and utilize advanced technologies in practical settings [17][19][68]. Group 4: Industrial AI and Future Prospects - Industrial AI is gaining traction, with a focus on enhancing efficiency and flexibility in manufacturing processes, as evidenced by the introduction of systems like the Industrial Copilot, which can improve production efficiency by up to 50% [39][41][36]. - The article suggests that the future of AI will involve deeper integration into various industries, emphasizing the importance of collaboration and open-source initiatives to bridge the technological divide [28][30][31]. Group 5: Global Collaboration and Governance - The Chinese government proposed the establishment of a global AI cooperation organization, aiming to foster international collaboration and address ethical and governance challenges in AI development [30][31]. - The conference underscored the need for a unified approach to AI governance, promoting standards and ethical guidelines to ensure responsible AI deployment [47][48].

Group 1 - The 2025 World Artificial Intelligence Conference in Shanghai showcased the innovative strength of Hangzhou companies, highlighting their advancements in various AI sectors including large models, intelligent robots, brain-machine interfaces, and AI terminals [1][2] - The event featured a record scale with over 70,000 square meters of exhibition space, more than 800 top global tech companies, and over 3,000 cutting-edge exhibits [1] - Hangzhou companies, including Alibaba, Yushun Technology, and Qiangnao Technology, formed a "Hangzhou Alliance" that demonstrated China's AI innovation capabilities on a global stage [1][2] Group 2 - In the field of large models, Alibaba Cloud's "Bailian" was recognized as a key exhibit, offering over 200 mainstream models for application development [2] - Ant Financial introduced the AI financial model "Agentar-Fin-R1," designed to assist users in stock trading and financial management [2] - The focus of the conference shifted from model development to application depth, with significant presentations in brain-machine interface technology from companies like Qiangnao Technology and Jialiang Medical [2] Group 3 - Alibaba's AI terminal product, Quark AI glasses, gained attention for its integration with Alibaba and Alipay ecosystems, supporting navigation and payment functions [3] - Other notable exhibitors from Hangzhou included Geely Auto and Xinhua San, contributing to a strong representation of the city's AI industry [3] - Hangzhou's continuous technological innovation solidifies its important position in the global AI landscape, showcasing the city's collective innovation strength and China's international competitiveness in AI [3]

Investment Rating - The report maintains a "Recommended" investment rating for the computer industry [2][11]. Core Insights - The WAIC 2025 showcased significant advancements in domestic AI technology, indicating a shift from being a "follower" to a "leader" in the AI sector. This includes developments in domestic computing power, large models, and AI applications, which are expected to evolve in a synergistic manner [11]. - Huawei's unveiling of the Ascend 384 SuperNode at WAIC 2025 represents a major leap in domestic AI computing capabilities, with over 80 mainstream large models already adapted for this technology. This architecture is anticipated to enhance the competitiveness of domestic computing power [7][11]. - The launch of the new 321 B-MoE large model by Jieyue Xingchen is set to open a new round of competition in multi-modal models, emphasizing the trend of "open-source + extreme inference efficiency" in domestic large model iterations [7][11]. - Alibaba Cloud's Baolian platform, recognized at WAIC 2025, has integrated over 200 mainstream models and attracted more than 200,000 developers, indicating a rapid acceleration in the commercialization of domestic AI applications [7][11]. Summary by Sections Domestic Computing Power - The Ascend 384 SuperNode features a high-bandwidth, low-latency interconnection among 384 NPUs, addressing communication bottlenecks within clusters and enhancing performance for model training and inference [7]. - The performance of SGLang and DeepSeek on the CloudMatrix384 has surpassed their performance on NV H100 and H800, showcasing the potential of domestic computing architectures [7]. Large Models and AI Applications - The 321 B-MoE model is expected to achieve three times the inference efficiency of DeepSeek-R1 on domestic chips and a 70% throughput increase on NVIDIA Hopper, highlighting the competitive edge of domestic models [7]. - The trend of open-source large models combined with domestic chips is projected to accelerate the growth of AI applications in China [7]. Key Companies to Watch - The report suggests focusing on companies such as Hengwei Technology (603496.SH), Youke De-W (688158.SH), YunSai ZhiLian (600602.SH), and Data Port (603881.SH) for domestic computing power. For large models and AI applications, companies like Dingjie Zhizhi (300378.SZ), HanDe Information (300170.SZ), and SuoChen Technology (688507.SH) are highlighted [11].

Group 1 - The 2025 World Artificial Intelligence Conference (WAIC) opened in Shanghai, showcasing cutting-edge AI achievements from top global tech companies [1] - The "Treasure of the Museum" award was given to eight notable AI products, including Huawei's Ascend 384 super node and Ant Technology's AI Health Manager AQ [1][2] - Over 600 companies participated in the exhibition, presenting more than 3,000 exhibits, with only eight items recognized as "Treasures of the Museum" [1] Group 2 - Ant Technology's AI Health Manager AQ has served over 100 million users since its trial run in September last year, providing health advice and expert-like diagnostic capabilities [2] - iFlytek's AI Learning Machine, another "Treasure of the Museum," offers personalized learning experiences for students across all educational stages [2] - Baidu's new digital human technology, NOVA, achieved a GMV of 55 million in a live stream and is set to be available to the public in October, enabling users to replicate professional live-streaming capabilities [2][3] Group 3 - Huawei showcased the Ascend 384 super node for the first time, highlighting its innovative computing capabilities and open-source hardware and software ecosystem [3] - Kuaishou's AI-generated video technology received an innovation award, while other companies were recognized for their advancements in AI models and applications [3]

Group 1 - AI Agents are recognized as a key driver for industrial intelligence upgrades, showcasing significant potential for practical applications [1] - Alibaba's latest open-source models achieved global championships in foundational, programming, and reasoning categories during the "2025 World Artificial Intelligence Conference" [1] - The AI programming model Qwen3-Coder has made substantial advancements in code capabilities, language understanding, and agent invocation [1] Group 2 - Alibaba Cloud's VP, Ye Jieping, announced that Tongyi Qianwen has achieved full-scale, multi-modal open-source status, breaking down barriers between open-source and closed-source models [2] - The AI model technology is undergoing a historic shift from incremental innovation to exponential leaps, with significant advancements noted since 2019 [2] - Alibaba plans to invest 380 billion yuan in the next three years for cloud and AI hardware infrastructure, exceeding the total investment of the past decade [2] Group 3 - Following the open-source release of Alibaba's three major models, the API call volume for Tongyi Qianwen surged, surpassing 100 billion tokens in just three days, outpacing models like GPT, Gemini, and Claude [2] - Tongyi Qianwen's download volume has exceeded 400 million globally, with over 140,000 derivative models created [2] - Alibaba Cloud's "Bailian" platform has over 200,000 developers working on more than 700,000 agents, showcasing a comprehensive "full-stack AI" capability [2] Group 4 - Ye Jieping emphasized that open-source has fostered a thriving ecosystem for large models, significantly enhancing their understanding, reasoning, and generation capabilities [3] - The "Agent-first" era is anticipated to fully emerge as AI transitions from dialogue to action [3] - Zhilian Recruitment's CTO shared insights on how AI Agents are being utilized for intelligent management in recruitment, enhancing efficiency through the use of open-source models [3] Group 5 - Zhilian Recruitment is collaborating with top domestic large model vendors to explore applications in human resources, addressing core efficiency pain points in recruitment processes [3] - The demand for AI-related job positions, such as data annotators and algorithm engineers, has increased year-on-year, highlighting the growing importance of digital and soft skills [3] - Future organizational structures are expected to diversify, with the emergence of AI employees necessitating a rethinking of management practices [3]

Core Viewpoint - The 2025 World Artificial Intelligence Conference (WAIC) in Shanghai showcases the rapid integration of AI across various sectors, emphasizing the importance of global cooperation in AI governance and innovation [2][3][4]. Group 1: Event Overview - The WAIC commenced on July 26, 2025, at the Shanghai Expo Center, featuring over 800 participating companies and an exhibition area exceeding 70,000 square meters [2][3]. - The event's theme, "Intelligent Era, Shared Future," highlights the collaborative potential of AI technologies across different industries [3]. Group 2: AI Applications and Innovations - AI applications are being integrated into diverse fields such as healthcare, finance, education, and tourism, with new products capturing significant attention at the conference [4][9]. - Notable exhibits include the Ant Group's AI Health Manager, which connects users with over 269 specialist AI agents and nearly 5,000 hospitals, serving over 100 million users since its launch [12]. Group 3: Key Technologies and Products - Major tech companies like Google, Amazon, Huawei, and Alibaba showcased their latest innovations, with products like Huawei's Ascend 384 supernodes and Alibaba's Quark AI glasses being highlighted as significant advancements [12][15]. - The Quark AI glasses, which feature capabilities such as communication, music, translation, and meeting notes, are positioned as a crucial development in wearable technology [16][19]. Group 4: Global AI Governance - The establishment of the Global AI Innovation Governance Center aims to foster international collaboration in AI governance, focusing on building consensus and standards [13]. - Geoffrey Hinton, known as the "father of AI," emphasized the need for global cooperation to prevent AI from surpassing human control, advocating for a community of AI safety research institutions [20][22].