2025 年 07 月 12 日 行业研究 评级：推荐(维持) [Table_Title] 智元入主上纬新材，xAI 发布大模型 Grok4 ——人形机器人行业周报 最近一年走势 | 行业相对表现 | | 2025/07/11 | | | --- | --- | --- | --- | | 表现 | 1M | 3M | 12M | | 电力设备 | 7.4% | 14.0% | 27.8% | | 沪深 300 | 3.1% | 7.0% | 15.8% | 相关报告 《电力设备行业周报：工信部召开光伏企业座谈 会，海风政策催化不断落地（推荐）*电力设备*李 航，邱迪，王刚，严语韬，李昂》——2025-07-06 《人形机器人行业周报：湖北省人形机器人产业投 资母基金设立，步科发布第四代无框力矩电机及中 空驱动器（推荐）*电力设备*李航，邱迪，李昂， 李铭全》——2025-07-05 | 研究所： | | | --- | --- | | 证券分析师： | 李航 S0350521120006 | | | lih11@ghzq.com.cn | | 证券分析师： | 邱迪 S0350522010002 | | | ...

Hugging Face’s new Reachy Mini doesn’t do much out of the box, and that’s kind of the point.It might be the Seinfeld of AI hardware, but that didn’t stop it from pulling impressive day-one sales.@EquityPod is talking open source bots, LangChain’s $1B pivot, Rivian’s new side hustle, and another weird week for X on today's episode 🤖https://t.co/auEEyzoHDG ...

henry 发自 凹非寺 量子位 | 公众号 QbitAI AI圈的"GitHub"居然开始卖机器人了！ HuggingFace最新发布开源桌面机器人ReachyMini，5小时内卖出超13万欧元（约为人民币109万元）！ 有线版本Reachy Mini Lite售价 299美元 ，无线版本Reachy Mini售价 499美元 。 换算下来，用一块Apple watch的钱你就可以把它带回家！ 评论区网友纷纷表示，马上下单！ Reachy mini有何不同？ Reachy mini身高28cm，体重1.5kg，拥有可移动的头部和能够旋转的身体。 脑袋上还有两根摇摇晃晃的天线，小巧的体型完全可以让你把它放在你的电脑旁边。 Reachy Mini虽小，却"麻雀虽小，五脏俱全"，在结构设计和AI接入上具备完整的系统框架。 它拥有六自由度的头部移动、全身旋转、动画天线、广角摄像头、多个麦克风和一个5瓦扬声器。 无线版配有RaspberryPi5作为计算核心，支持Wi-Fi和电池供电，麦克风数量也从Lite版的2个提升到4个，并额外加入了加速度计，使其具备 更完整的传感与交互能力。 相比之下，Lite版则需外接电脑运 ...

Group 1 - Musk released Grok4, highlighting its superior performance in various tests, particularly in the "ultimate human exam" surpassing competitors [1] - Grok4's training approach has shifted to emphasize "first principles" thinking, learning to use tools to solve problems during the training phase [1] - Grok faces controversy over the "mechanical Hitler" issue, as its unfiltered approach attracts users but also raises concerns about AI alignment challenges [1] Group 2 - Microsoft open-sourced Phi-4-mini-flash-reasoning, utilizing the innovative SambaY architecture, achieving a 10x increase in reasoning efficiency and a 2-3x reduction in latency [2] - The SambaY architecture enables efficient memory sharing across layers without explicit positional encoding, significantly enhancing long context processing capabilities [2] - The new model is suitable for resource-constrained devices, running on a single GPU, excelling in advanced mathematical reasoning and long text generation, making it ideal for educational and research fields [2] Group 3 - Perplexity officially launched the AI browser Comet, centered around "agent search," competing with Google Chrome [3] - Comet's three main value propositions include personalized understanding of user thinking, powerful and user-friendly content comprehension, and efficiency improvements reducing tab switching [3] - Comet features rich functionalities, capable of replacing user actions on the web, intelligently processing content, managing email calendars, and searching personal data, currently supporting Mac and Windows systems [3] Group 4 - OpenAI completed the acquisition of io company, with former Apple designer Jony Ive and his team LoveFrom joining to take on deep design and creative responsibilities [4][5] - Ive is expected to assist OpenAI in developing new intelligent hardware products, with initial ideas being transformed into feasible designs [5] - The io company, co-founded by Ive and several experts, includes hardware and software engineers and scientists, and will closely collaborate with OpenAI's R&D team [5] Group 5 - Google released new medical AI models: the multimodal MedGemma 27B and the lightweight encoder MedSigLIP, expanding the HAI-DEF medical model collection [6] - The MedGemma series includes 4B and 27B versions, supporting image and text input with text output; the 4B version achieved a 64.4% accuracy rate in medical Q&A tests, while the 27B version reached 87.7% [6] - MedSigLIP, with only 400 million parameters, is a medical image encoder optimized through various medical imaging techniques, suitable for image classification, zero-shot classification, and semantic retrieval, providing visual understanding for MedGemma [6] Group 6 - Tencent launched a co-creation activity for the 2026 "Year of the Horse" zodiac penguin, with requests surging 300% within hours and token usage doubling, prompting urgent server expansion [7] - The activity invites users to design the 2026 "Horse Goose" figurine using the Mix Yuan 3D AI creation engine, allowing text input, image uploads, or sketch submissions to generate designs [7] - Outstanding works will have the opportunity to be co-branded with Tencent for mass production and sold in official merchandise stores, with the activity closing on July 27, 2025 [7] Group 7 - OpenAI plans to release an "open weight model," similar to the o3 mini level, as early as next week, allowing companies to deploy it themselves, marking the first model weight release since 2019 [8] - OpenAI is developing an AI browser based on Chromium, which will process web content within the ChatGPT native interface, enabling AI agents to execute tasks directly, challenging Google Chrome [8] - OpenAI is expanding its business scope from model development to browsers and other user interfaces, indicating its ambition for technological leadership and ecosystem control [8] Group 8 - Hugging Face and Pollen Robotics jointly launched the open-source robot Reachy Mini, starting at $299, designed for human-robot interaction and AI experimentation [10] - Reachy Mini offers a basic version ($299) and a wireless version ($449), supporting Python programming and equipped with multimodal interaction features like cameras, microphones, and speakers [10] - The robot stands 28 cm tall, weighs 1.5 kg, provides 15 preset behaviors, is fully open-source and extensible, with the basic version expected to ship by late summer 2025 and the wireless version in batches starting fall 2025 [10] Group 9 - Meta released a 40-page report, positioning the "mental world model" alongside the physical world model as a key component of embodied intelligence [11] - The mental world model focuses on human goals, intentions, emotional states, social relationships, and communication methods, enabling AI to understand human psychological states and engage in social interactions [11] - Meta proposed a dual-system architecture integrating "observational learning" (System A) and "action learning" (System B), where the former provides abstract knowledge and the latter explores actions for more efficient agent learning [11] Group 10 - Top AI products like Cursor, Perplexity, and Lovable have adopted a "anti-framework" approach, building directly on basic AI units rather than using frameworks [12] - Frameworks have become innovation barriers in the rapidly changing AI field, leading to excessive abstraction, bloated structures, and slow iterations, while basic units offer combinability and specialization [12] - The basic unit method (e.g., Memory, Thread, Tools) allows developers to construct AI products like building blocks, reducing cognitive load and enhancing performance and flexibility, better suited for rapid AI technology iterations [12]

1、 国产四足机器人"黑豹2.0"破世界纪录 近日，"智绘光谷"CMG世界机器人技能大赛上，国产四足机器人"黑豹2.0"以10.3米/秒速度打破世界纪 录。这标志着国产四足机器人在速度上达到国际领先水平，增强了技术自信与创新动力。近年来，国产四 足机器人应用场景不断拓展，涵盖特种作业、灾难救援、安防、家用等多个领域。面对这一市场，A股已 有近30家上市公司布局四足机器人产业链，在硬件供应、技术服务、整机制造等方面积极行动，有望带动 行业整体技术提升，推动商业化进程。 4、 极智嘉港交所成功上市，成"全球AMR仓储机器人第一股" 2、 智能手术机器人SRT-H自主完成胆囊切除 约翰斯·霍普金斯大学主导研发的智能手术机器人SRT-H，近日首次在患者模型上成功自主完成胆囊切除手 术，展现出了与熟练人类外科医生相当的专业能力。该机器人基于机器学习架构构建，能实时适应个体解 剖结构、即时决策，并从互动中学习优化。此次手术实现了100%成功率，尽管耗时略长，但结果与专业 医生相当。这一进展标志着手术机器人从"执行任务"迈向"理解手术"，为临床实际应用打下基础，有望缓 解医生资源紧张问题。 3、 湖北22家鄂企携人形机器人 ...

｜2025年7月10日星期四｜ NO.1英伟达成为全球首家市值突破4万亿美元公司 当地时间7月9日，英伟达市值短暂突破4万亿美元，成为全球首家达到这一里程碑的上市公司。受AI技 术需求持续激增的推动，英伟达的股价一度上涨2.5%，达到164.42美元/股的历史高点。最终英伟达收 于162.88美元/股，市值达到3.97万亿美元。 点评：英伟达的表现或成为科技股信心的风向标。这一里程碑事件，可能引发市场对科技龙头股的关 注。 NO.2X CEO琳达·亚卡里诺宣布离职 当地时间7月9日，Hugging Face宣布，其 Reachy Mini 桌面机器人现已开始接受订单。Reachy Mini有两 个版本：第一个版本为售价499美元的无线版本Reachy Mini Wireless；第二个版本是需要连接到计算设 备的Reachy Mini Lite，价格为299美元。该机器人预装了一组演示程序，并与开源机器学习平台 Hugging Face Hub集成。任何人都可以为Reachy Mini构建自己的特定功能和应用程序，然后与社区分 享。 点评：Hugging Face推出Reachy Mini桌面机器人并开放订购 ...

2.【Hugging Face开源小参数模型SmolLM3】北京时间7月9日凌晨，Hugging Face首席执行官克莱门特·德朗 格（Clement Delangue）宣布，Hugging Face发布并开源小参数模型SmolLM3。拥有128k上下文窗口；支持 英语、法语、西班牙语、德语等6种语言；支持深度思考和非思考双推理模式。（财联社） 3.【浙江大学发布高精准基因组设计AI模型】浙江大学郭国骥教授团队开发出一款用于基因组预测设计的深度学 习AI模型"女娲CE"，能够以超过90%的准确率预测基因组调控区域发生突变之后带来的表型变化，并结合疾病 表型设计出相应的治疗位点。相关成果已发表于国际学术期刊《细胞》。 （财联社） 4.【Hugging Face 桌面机器人 Reachy Mini 开订：长相呆萌，支持超 170 万个 AI 模型】据外媒TechCrunch 报道，Hugging Face旗下最新桌面机器人Reachy Mini的订单现已正式开放，开发者现在已可动手组装与测试。 Reachy Mini将推出两个版本。无线版名为Reachy Mini Wireless，内置Raspberry 5 微 ...

Hugging Face opens up orders for its Reachy Mini desktop robots | TechCrunch https://t.co/VnfiBSiyT4 ...

Core Insights - Hugging Face has officially entered the embodied AI sector by launching two humanoid robots, HopeJR and Reachy Mini, marking its first venture into hardware after establishing a strong presence in software [2][3] - The acquisition of Pollen Robotics, the creator of the modular humanoid robot Reachy, has provided the technological foundation for this transition, allowing Hugging Face to leverage Pollen's engineering and design principles in its new robots [2][3] - The company aims to expand beyond chatbots and model hosting to create AI systems capable of interacting in real-world environments while maintaining its commitment to open-source principles [2][3] Company Overview - Hugging Face, founded in 2016, initially launched a chatbot application for teenagers before evolving into a comprehensive machine learning development platform that supports model hosting, dataset sharing, and API integration [3] - The company recognizes the necessity for AI to interact with the physical world, prompting its focus on embodied intelligence and robotics, which are designed for experimentation, collaboration, and reproducibility [3] Product Details - HopeJR is a full-sized humanoid robot designed for developers and researchers, featuring a modular platform with multiple cameras, articulated limbs, and voice interaction capabilities, allowing for testing of language models and perception systems in real-world scenarios [6] - HopeJR boasts 66 degrees of freedom (DOF) and aims to achieve walking capabilities in real-world environments following successful simulations [6] - Reachy Mini is a smaller version of the original Reachy robot, designed for educational purposes and independent developers, featuring functionalities like expression display and basic object manipulation [9] - The introduction of Reachy Mini aims to lower barriers for developers, making it easier to experiment without the costs and complexities associated with larger robots [9] Future Outlook - While specific pricing and release dates for HopeJR and Reachy Mini have not been disclosed, both robots will operate under a loose open-source license [10] - The company plans to introduce benchmark challenges, developer toolkits, and a dedicated robot section on Hugging Face Hub, along with deep integration with the PyTorch robotics platform LeRobot [10] - Hugging Face's CEO emphasizes the open-source strategy to prevent monopolization of robotics technology by a few large companies, reiterating the mission to democratize AI and robotics [10]

生成式AI 一、 Mamba作者之一 提出2种专为推理量身定制的注意力机制 1. Mamba核心作者提出两种专为推理定制的注意力机制GTA和GLA，可将解码速度和吞吐量 最高提升2倍； 2. GTA是GQA的替代品，能减少约50%KV缓存使用，而GLA比DeepSeek使用的MLA解码 速度更快； 3. 这些机制通过优化内存使用和计算逻辑，在不牺牲模型生成质量前提下提升大语言模型推 理效率。 https://mp.weixin.qq.com/s/61k_W7m21hWBYv2Er00E9A 2. 其核心优势包括出色的角色一致性保持、快速生成速度(10秒左右)和良好的风格迁移能 力，可用于物体修改、背景替换等多种场景； 3. 目前存在的局限包括多轮编辑后质量退化、无法参考多图像、改变元素过多会报错，使用 时应遵循"明确保留内容"的提示词原则。 https://mp.weixin.qq.com/s/iLt00rVtArp9fW95X0NTmA 二、 Flowith的Agent Neo已全面开放， 首个无限执行 Agent 1. Flowith推出Agent Neo，声称是全球首个支持无限执行、无限产出的AI Ag ...