Core Viewpoint - Mech-Mind showcased its self-developed general-purpose robot "Eye-Brain-Hand" technology at WAIC 2025, featuring nearly 10 cutting-edge applications in embodied intelligence, including dual-arm folding clothes, humanoid robot picking, and sorting a vast array of objects [2][4]. Group 1: Technology and Applications - The "Eye-Brain-Hand" technology integrates various robotic forms, including single-arm, dual-arm, and humanoid robots, applicable across multiple industries and tasks such as smart manufacturing, logistics, pharmaceuticals, retail, food, and household services [4]. - The dual-arm robot, equipped with the "Eye-Brain" system, possesses human-like reasoning, planning, and decision-making capabilities, enabling it to autonomously execute complex tasks like folding clothes efficiently [6]. - The dual-arm robot can autonomously sort a vast array of random objects using a self-developed multimodal large model, allowing it to recognize and classify items in real-time, meeting the high-speed sorting demands in industries like logistics and e-commerce [8]. - The humanoid robot, based on the DARWIN platform, can autonomously perform shelf picking tasks, demonstrating high mobility and adaptability in narrow spaces [10]. - The single-arm robot integrates advanced technologies to understand natural language commands and execute various complex tasks autonomously [12]. Group 2: Performance Metrics - The robot can achieve a sorting speed of up to 1600 items per hour, widely used in industrial, logistics, e-commerce, and retail sectors [14]. - The robot can accurately identify transparent objects with sub-millimeter precision, suitable for high-speed sorting in pharmaceutical and industrial applications [16]. - Utilizing advanced imaging and recognition technology, the robot can identify complex metal parts from over 3 meters away with a visual accuracy of ±0.5mm, successfully implemented in the automotive industry [18]. - The robot can quickly complete small bolt feeding tasks with a grabbing speed of less than 1.9 seconds per item, applicable in automotive and home appliance sectors [20]. Group 3: Technological Innovations - Mech-Mind also showcased its self-developed Mech-Eye high-precision 3D camera, capable of micron-level precision and fast scanning speeds, providing robots with enhanced 3D vision capabilities [21]. - The Mech-GPT multimodal large model equips robots with human-like learning, understanding, planning, and decision-making abilities, allowing them to execute tasks based on natural language instructions [22]. - The Mech-Hand five-finger dexterous hand design allows for efficient and precise manipulation of a wide variety of items, overcoming the limitations of traditional gripper applications [23]. Group 4: Market Impact - The advanced technology and diverse applications showcased by Mech-Mind made its booth one of the most popular at WAIC 2025, highlighting the company's strong market presence and innovation in the field of robotics [27].

Virtual Agent通过方言理解、全渠道整合及弹性架构，为出海企业提供可复制的智能运营 范式。 本文为IPO早知道原创 作者｜ 苏打 微信公众号｜ipozaozhidao "2022年，大家认为ChatGPT只是一个产品。2024年，企业级应用开始发力，厂商看到了一些业务 价值。今年，DeepSeek让大模型的ToB变得平民化，Manus也让大家看到Agent的新价值主张。这 些都带来了新启发"。 据 IPO早知道消息， 7月28日 ， 容联云在 2025 WAIC世界人工智能大会期间，承办"重塑升力 ·AI智启新增长"AI A GENT 落地实践与产业创新论坛。论坛聚焦 AI Agent的技术演进与产业融 合，通过垂直场景解决方案与全球化趋势洞察 ，揭示 AI Agent如何重构企业增长范式，推动新质生 产力发展。 容联云副总裁、诸葛智能创始人孔淼 现场详细分享了容联云的最新 AI进展及动态。 他表 示，容联云始终以技术前瞻性为驱动，在每一次迭代中走在行业前沿 ——从LLM Feature（大 模型功能嵌入）到Workflow（流程/编排级应用）到Agent（自主规划与工具调用），并 将大模型 真正融 ...

Core Viewpoint - The article highlights the impressive product showcase of Magic Atom at WAIC 2025, featuring the new humanoid robot MagicBot Z1 and the wheeled quadruped robot MagicDog W, emphasizing their advanced capabilities and potential applications in various industries [2][8]. Product Showcase - MagicBot Z1 features 24 basic degrees of freedom, expandable to 50, with a maximum joint torque exceeding 130 N·m, enabling complex movements [2] - MagicDog W, launched shortly after MagicBot Z1, boasts 17 degrees of freedom and utilizes a "wheel-foot composite" movement mode, achieving speeds over 3 m/s on flat surfaces and adapting to obstacles with a stable four-legged gait [4][5]. Interaction and Usability - MagicBot Z1 showcases dynamic performance with actions like "getting up" and "bending," and offers interactive experiences with audiences through its visual and tactile perception systems, making it a responsive companion [3] - The humanoid robot "Xiao Mai" demonstrated its operational capabilities in industrial settings, performing tasks like "transporting and gluing" under dynamic conditions, highlighting its stability and adaptability [7]. Application Versatility - Magic Atom's products are designed for diverse applications, including industrial, commercial, and educational sectors, showcasing their versatility and technological strength [8] - The humanoid robot "Xiao Mai" serves as an intelligent guide, capable of voice interaction and action recognition, enhancing user experience in various environments such as museums and corporate exhibitions [7][8].

Core Viewpoint - The article emphasizes the innovative development of the E20 eVTOL by the company, showcasing its potential to provide accessible air travel solutions globally through advanced technology and AI integration [5][6]. Group 1: E20 eVTOL Overview - The E20 eVTOL is a manned electric vertical take-off and landing aircraft developed by the company, featuring advanced flight control technology and AI systems [5]. - It has a maximum speed of 320 km/h, can carry 5 passengers, and has a design range of 200 km, with an operational cost of 3-4 yuan per kilometer [5]. Group 2: Technological Innovations - The intelligent avionics and flight control system of the E20 eVTOL is co-developed with the Aviation Industry Corporation of China, incorporating AI algorithms for enhanced precision and stability during complex flight phases [5]. Group 3: Industry Development - Shanghai is actively building a "Sky City" and developing a low-altitude economic industry system, focusing on a network of facilities, air routes, and services to create a national-level low-altitude economic demonstration zone [5]. - The city aims to accelerate technology transfer and commercial implementation through AI empowerment and collaborative efforts in the Yangtze River Delta region [5].

Core Viewpoint - The article discusses the launch of a comprehensive AI efficiency enhancement plan by Wu Wen Xin Qiong, aiming to meet "infinite" demands with limited resources through three core products designed for various computing power scenarios [2][3]. Group 1: Product Offerings - Wu Wen Xin Qiong introduced three main products: "Wu Qiong AI Cloud" for global computing networks ranging from 10,000 to 100,000 cards, "Wu Jie Intelligent Computing Platform" for large-scale intelligent computing clusters from 100 to 1,000 cards, and "Wu Yin Terminal Intelligence" for limited computing terminals from 1 to 10 cards [2]. - The overall solution is a soft-hard collaborative system for future intelligent infrastructure, providing full-link support from model scheduling to application deployment across diverse computing power scenarios [2]. Group 2: Efficiency and Resource Management - The CEO highlighted the challenge of limited resources in the pursuit of Artificial General Intelligence (AGI), emphasizing the need to enhance intelligent efficiency and expand computing resources [4][5]. - The article notes that the evolution of AI from traditional algorithms to AI 1.0 and AI 2.0 has been driven by the Scaling Law, which continuously pushes the boundaries of intelligence through resource utilization [4]. - Wu Wen Xin Qiong aims to maximize the potential of every unit of computing power, asserting that the intersection of improved efficiency and resource expansion is where the new AGI threshold lies [5]. Group 3: Value Space of AI Applications - The value space for AI application deployment is defined as the product of scenario scale, computing resources, and intelligent efficiency [5]. - The company is committed to achieving the ultimate vision of "limitless intelligence and precise computing" by harmonizing scenario scale, computing resources, and intelligent efficiency [5].

Core Viewpoint - Wallrun Technology has once again won the SAIL Award, following its previous win in 2022, highlighting its continued innovation in the field of AI and computing technology [1][2][3]. Group 1: Award Recognition - The "Distributed OCS All-Optical Interconnect Chip and Super Node Application Innovation Solution," co-developed by Xizhi Technology, Wallrun Technology, and ZTE, received the WAIC 2025 SAIL Award [2]. - This marks Wallrun Technology's second consecutive win of the SAIL Award, emphasizing its leadership in AI technology [3]. Group 2: Product Launch - On July 28, the LightSphere X, the first optical interconnect GPU super node in China, was officially launched by Shanghai Yidian in collaboration with Xizhi Technology, Wallrun Technology, and ZTE [3]. - The LightSphere X utilizes Xizhi Technology's innovative distributed optical switching technology and Wallrun Technology's proprietary architecture for high-performance GPU liquid cooling modules [3]. Group 3: Technological Advancements - The rise of trillion-parameter large models and multimodal training is driving computing clusters into the "ten-thousand-card collaboration" era, with super node architecture being a key technological path [4]. - LightSphere X employs optical interconnect technology to overcome traditional physical limitations of super nodes, allowing for elastic expansion and reduced deployment costs [4][5]. - The distributed optical circuit switch (dOCS) technology enhances the flexibility and scalability of the super node, allowing for real-time topology reconstruction and dynamic adjustment of super node scale based on model computing needs [5]. Group 4: Performance and Scalability - The LightSphere X features a powerful GPU module designed with multi-chiplet and CoWoS 2.5D packaging, significantly enhancing cluster training and inference performance [6]. - The unified management platform for the intelligent computing cluster enables intelligent control over the entire lifecycle of the super node, improving training performance and ensuring stable model operation [6]. Group 5: Ecosystem Development - LightSphere X aims to build a fully autonomous intelligent computing ecosystem, providing robust support for China's AI development [7]. - The hardware layer is designed to be open and compatible, eliminating risks of ecosystem lock-in and ensuring independence from advanced semiconductor manufacturing processes [8]. - The collaborative system of "open protocols + autonomous technology + self-developed software" creates a closed-loop of self-controllable technology and ecosystem [9].

Core Viewpoint - The article discusses the launch of GLM-4.5, an open-source state-of-the-art model designed for reasoning, coding, and agent applications, marking a significant technological breakthrough for the company [2]. Group 1: Model Overview - GLM-4.5 integrates reasoning, coding, and agent capabilities into a single model, achieving state-of-the-art performance in open-source [3]. - The model features a mixed expert (MoE) architecture with a total parameter count of 355 billion and 32 billion active parameters for GLM-4.5, while GLM-4.5-Air has 106 billion total parameters and 12 billion active parameters [3]. - GLM-4.5 offers two modes: a thinking mode for complex reasoning and tool usage, and a non-thinking mode for immediate responses [3]. Group 2: Performance Metrics - The model was evaluated against 12 representative benchmarks, achieving the third-best global score, the first among domestic models, and the top position among open-source models [4]. - GLM-4.5 has half the parameter count of DeepSeek-R1 and one-third of Kimi-K2, yet outperforms them in several benchmark tests due to higher parameter efficiency [6]. Group 3: Cost and Efficiency - GLM-4.5 demonstrates significant cost and efficiency improvements, with API call prices as low as 0.8 yuan per million tokens for input and 2 yuan per million tokens for output [8]. - The model was tested against Claude Code, Claude-4-Sonnet, Kimi-K2, and Qwen3-Coder across 52 programming tasks, showing competitive advantages, particularly in tool invocation reliability and task completion rates [10]. Group 4: Broader Implications - In July, the company also released GLM-4.1V-Thinking, a new generation multimodal model that excels in reasoning capabilities, achieving top scores in 23 out of 28 authoritative evaluations [11]. - The company is rapidly expanding its overseas business, providing infrastructure solutions to various countries in Southeast Asia, the Middle East, and Africa [13]. - The company aims to establish a "verifiable, responsible, and standardized" technological image in emerging markets ahead of its Western competitors [14].

Core Viewpoint - The article emphasizes that the field of embodied intelligence is currently one of the hottest subfields in the AI market, with significant technological advancements and investment opportunities emerging rapidly [4][9]. Group 1: Company Overview - It Shihang, founded by Dr. Chen Yilun and Li Zhenyu, is a leading player in the field of embodied intelligence and autonomous driving, with a team that has extensive experience in the industry [3][4]. - The company has completed two rounds of financing totaling over $240 million within six months, with notable investors including Qiming Venture Partners, BlueRun Ventures, and Meituan [4]. Group 2: Technological Advancements - Dr. Chen highlighted four major changes in embodied intelligence technology over the past year: 1. Whole body control has fully entered the AI era 2. End-to-end systems for autonomous driving are now fully operational at the product level 3. Multi-modal large models still have significant growth potential due to unsaturated data 4. High-degree-of-freedom operation terminals, such as dexterous hands, are rapidly iterating and show mass production potential [5][9]. - The top-level technology stack for embodied intelligence is converging, but there are still six major areas of consensus and disagreement within the industry [10]. Group 3: Market Strategy - It Shihang follows a "golden triangle" logic in its market strategy, focusing on high-value, scalable, and challenging problems that previous generations of robotic technology have struggled to solve [12]. - The company encourages talent from various fields, including traditional robotics, autonomous driving, and multi-modal large models, to join and contribute to the exciting developments in embodied intelligence [12]. Group 4: Relationship with Autonomous Driving - Autonomous driving is considered a key sub-task of embodied intelligence, representing mobility and navigation capabilities, with significant value in end-to-end systems [6][13]. - Concepts from autonomous driving, such as defining AI in 4D space and the boundaries of imitation and reinforcement learning, can be applied to the field of embodied intelligence [13].

Core Viewpoint - The article discusses the strategic launch of edge computing by Aixin Yuan Zhi, emphasizing the importance of AI transitioning from technology to product, and the necessity of a cloud-edge-end infrastructure to overcome challenges in cost, real-time processing, and data privacy [2][4]. Group 1: Company Strategy and Vision - Aixin Yuan Zhi aims to provide high cost-performance AI chips and reduce technical development barriers for partners through standardized development tools and algorithm optimization frameworks [2][4]. - The company is focused on creating an "Edge Intelligence Community" to promote the development of edge computing and facilitate the widespread application of intelligent technology across various industries [2][4]. - The company has officially launched its edge computing business line, marking a new journey towards edge intelligence [4]. Group 2: Industry Collaboration and Innovation - Collaboration with partners like Wenshin Qiong and M5Stack is highlighted, focusing on solving edge-side inference challenges and providing commercial support for vertical scenarios [3][4]. - The partnership with Guanghe Tong emphasizes the need for AI solutions to be tailored to industry demands, showcasing successful implementations in areas like unmanned retail [3]. - The collaboration with Lingjing Acoustics aims to create a panoramic sound industry model, addressing content scarcity and scene adaptation issues in various applications [4]. Group 3: Future Outlook and Trends - The article suggests that soft-hard collaboration will be crucial for the advancement of edge intelligence, particularly in human-machine interaction scenarios such as healthcare and robotics [2][3]. - The ongoing innovation in business models is seen as a driving force for both technological and commercial growth, with companies like M5Stack evolving to provide efficient IoT development solutions [3].

当硬件性能与软件效率形成合力，具身智能的边界将被不断拓宽。 本文为IPO早知道原创 作者｜ Stone Jin 微信公众号｜ipozaozhidao 据 IPO 早 知 道 消 息 ， 跨维智能 于 7 月 27 日 在 WAIC 2025 正 式 发 布 了 全 新 一 代 人 形 机 器 人 DexForce W1 Pro —— 作为 W1 的迭代升级款，其在机器人本体软硬件算法与开发工具链上实现 全方位革新，面向各个行业的用户提供丰富的硬件配置选择与 EmbodiChain 具身智能开发平台。 这里需要指出的一点是， 不同于停留在 "实验室环境"的基于舵机方案的实验用机器人，或是主要服 务于"情绪价值"的基于准直驱方案的展示用机器人，一个能创造真实价值的精巧操作型机器人，必 须在核心零部件、机身构型、电路设计及控制算法上实现深度耦合，从而达到产品级的高稳定性、高 精度、高柔性与长时运行能力。 DexForce W1 Pro 就 融合了 X-Wiz 提供的模块化多层次开发接口，并借助 EmbodiChain 赋予 的强大数据扩增与生成能力，能够在真实场景的作业需求牵引下，自主完成高精度、高稳定性的机器 人 ...