Summary of the Conference Call for UBTECH Company Overview - **Company**: UBTECH - **Technology**: Development of the Thinker large model for humanoid robots, showcasing advanced capabilities in task understanding, environmental perception, cognitive decision-making, and task planning [2][4] Key Points Industry and Technology Advancements 1. **Performance in International Evaluations**: The Thinker model has achieved 9 global firsts in international evaluations, demonstrating its technological leadership [2][4] 2. **Robust Execution in Industrial Applications**: The humanoid robots exhibit high efficiency and precision in tasks such as handling and sorting, aided by the integration of the Thinker model with Visual Object Analysis (VOA) technology [2][4][5] 3. **Self-Correction Capabilities**: The robots can autonomously adjust during operations, showcasing strong generalization abilities even in unforeseen circumstances [5][10] Model Development and Data Handling 1. **Data-Centric Approach**: The development has shifted from model-centric to data-centric, achieving nearly 100% automation in data annotation processes, emphasizing the importance of high-quality data [3][6] 2. **Diverse Data Sources**: The training data primarily consists of open-source data, supplemented by real machine and customer site data, effectively addressing hardware heterogeneity issues [5][6] 3. **Cost Control**: Utilizing open-source data significantly reduces research and development costs [5][6] Engineering and Optimization 1. **Efficient Model Deployment**: The Thinker model operates efficiently on limited computational resources through techniques like model distillation and quantization [5][6] 2. **Performance Improvement Factors**: Key factors include multi-modal data fusion, advanced algorithms, and engineering optimizations that enhance stability and reliability in real-world applications [6][10] Future Directions and Goals 1. **Upcoming Model Updates**: In the next 3-6 months, UBTECH aims to explore visual language action (VLA) and world model directions, with plans to release related results soon [7][10] 2. **Adaptability for Specific Models**: The current model serves as a foundational model, which will be fine-tuned for specific robot models like Work S2 [7][10] Collaboration and Open Source 1. **Support for Developers**: The open-source model allows developers to perform secondary fine-tuning and includes tools for hardware optimization, facilitating efficient deployment on UBTECH robots [9][10] 2. **Data Security Measures**: UBTECH ensures that sensitive information in training data is protected and complies with cross-border data transmission policies [9][10] Unique Selling Propositions 1. **Compact Model Size**: The Thinker model is notably small (4B), enhancing deployment efficiency compared to competitors [11][12] 2. **Real-World Data Utilization**: The reliance on real industrial scenario data increases the model's reliability and applicability in practical environments [11][12] Challenges and Solutions 1. **Long-Tail Problem Management**: UBTECH addresses the long-tail problem by focusing on failure cases, collecting and annotating them for model iteration, thus creating a data feedback loop to enhance overall performance [10][12] Conclusion UBTECH's advancements in humanoid robotics through the Thinker model highlight its commitment to innovation, efficiency, and practical application in industrial settings, positioning the company favorably within the robotics industry.

这五年，玄武经济总量连破3个百亿大关，2025年GDP达1439亿元，增速5.6%，位居主城第一，服务业 增加值占比超98%、位列全市第一。 这五年，玄武锚定"1+2+2"现代化都市型产业体系，抢先布局数据要素、大模型、绿色低碳和跨境电商 未来产业，文旅产业占GDP近20%，限上社零总额占全市17%、首次位列第一，全省60个国家备案大模 型占了22个，江苏国际数据港（603881）、国际数碳谷(GDC)、玄武大模型工厂初具雏形。 这五年，玄武系统谋划片区更新、产业焕新、功能革新。红山新城蝶变为"城市新中心"，徐庄高新区打 造高品质"生态硅谷"，生态质量指数位列全省第二，连续五年入选中国最美县域。 这五年，玄武让"好日子"可感可知。15分钟教育、医保、养老、生活等服务圈基本实现全覆盖，国企物 业托管住宅小区278个、服务居民超6万户，人均可支配收入超9.5万元，连续12年保持全省第一。 5年，1800多个昼夜，南京老城玄武的"大象转身"，日月可鉴。 打破空间束缚，摆脱要素制约，布局新兴赛道，更新城市（300778）界面，解码善治之道……过去五 年，玄武以新思想新理念为指引，紧扣特大城市中心城区高质量转型的发展主 ...

Group 1 - The core point of the article highlights that southbound funds have reduced their holdings in China Software International (00354.HK) by 868,000 shares on February 2, with a total net reduction of 1,302,000 shares over the past five trading days [1] - Over the last 20 trading days, there have been 12 days of net increases in southbound fund holdings, totaling 26,244,000 shares [1] - As of now, southbound funds hold 859 million shares of China Software International, accounting for 31.43% of the company's total issued ordinary shares [1] Group 2 - China Software International Limited is an investment holding company that provides global technology software and information technology services [1] - The company operates through two main segments: technology professional services and internet information technology services [1] - Its primary business focuses on the development of generative artificial intelligence (AIGC), sales of large model software and hardware, and digital transformation consulting services for enterprise resource planning (ERP) [1] - Key products include the "Question Series" solutions, large model application integrated machines, and the Lingxi AI application platform [1] - The company serves various sectors, including water conservancy, transportation, government platforms, military, energy, education, and finance, operating in both domestic and international markets [1]

（来源：智通财经） 上海市政协委员、上海商汤智能科技有限公司董事长、CEO徐立在谈及大模型发展时表示，在中美模型 创新方面，美国呈现出由超级明星主导的模式，中国在创新生态上如何构造超级个体、围绕这些人员构 建团队，是下一步创新的重要命题。他认为，大模型生产力工具发展到一定程度，构造的是端到端交付 能力，但仍需人和机器协同，呈现出生产力工具赋能超级个体的发展趋势。若大模型仅成为过程降本增 效的管理工具，商业化很难闭环；但一旦做成过程化交付、赋能个体完成交付，带来的生产力爆发非常 明显。 炒股就看金麒麟分析师研报，权威，专业，及时，全面，助您挖掘潜力主题机会！ ...

Core Viewpoint - The article discusses the competitive landscape of the "AI Red Packet" initiative during the Spring Festival, highlighting the strategic positioning of major tech companies like Tencent, Alibaba, Baidu, and ByteDance in leveraging AI to enhance user engagement and ecosystem integration [2][4]. Group 1: AI Red Packet Battle - The total amount of red packets distributed by major companies exceeds 4 billion yuan, marking a significant strategic move to enhance AI user engagement [2]. - Alibaba's "Spring Festival Treat Plan" involves a record investment of 3 billion yuan, while Tencent contributes 1 billion yuan through its "Yuanbao" initiative, and Baidu allocates 500 million yuan for its "Wenxin Assistant" [3]. - The competition is not just about monetary distribution but also about integrating AI into social interactions and commercial ecosystems [4]. Group 2: Market Reactions and Company Performance - Investor interest in AI applications has surged, with numerous inquiries about companies' involvement in the AI red packet initiatives, indicating a strong market response [3]. - As of February 2, 72 companies in the AI application sector have released performance forecasts, with 20 expecting growth, 12 turning profitable, and 25 continuing to report losses [5]. - Companies with strong technological foundations and clear application scenarios are beginning to see positive financial results, while those in transition face challenges [5]. Group 3: Future of AI Applications - The year 2026 is anticipated to be a pivotal moment for AI applications, with the current red packet initiatives serving as a critical turning point for commercializing AI technology [5]. - Companies like iFlytek are projecting significant profit growth, attributing success to the integration of AI models into existing solutions, which address customer pain points [6]. - The article emphasizes that not all companies will benefit from the AI boom, as many traditional firms may struggle to adapt to the evolving technological landscape [8].

【 策略 】大跌之后：良机就在眼前 【光伏设备 】SpaceX 申请百万卫星布局轨道 AI 数据中心，太空光伏与设备环节有望率先受益 大跌之后：市场有望逐步企稳，投资良机就在眼前。 近期中国市场波动较大，全A平均股价调整达到6%，上证指数跌至临近4000点，并在午后出现恐慌性抛 售。市场下跌由多因素耦合：1）预期层面，市场正快速评估新提名美联储主席凯文·沃什的政策主张，其"降息+缩表"的倾向引起金融条件紧缩的担心；2） 微观交易，黄金/白银等商品价格剧烈波动，以及前期A股权重ETF持续被抛售，叠加临近春节长假观望情绪升温，内外交织之下中国市场出现大跌。国泰海通 策略认为，大跌之后，市场有望逐步企稳，在春节前重拾上升势头，坚定看好中国股市：第一，全球市场正快速计入美联储潜在的鹰派货币立场，但从路径角 度沃什的降息立场更鸽派和确定；第二，中国政府政策重心正转向内需主导，并作为首要任务，有望提振中国经济前景与资产回报；第三，中国证监会近期再 次强调"全力巩固资本市场稳中向好势头"，并推出中长期资金入市新政。市场的无差别恐慌抛售令交易加快出清，投资中国的良机就在眼前。 市场快速计价美国"缩表"风险，但沃什的降息立场更 ...

（编辑 丛可心） 证券日报网讯 2月2日，天阳科技在互动平台回答投资者提问时表示，公司主要聚焦于人工智能、区块 链、云计算和大数据等新一代金融科技领域，为客户提供解决方案及服务，并在量子计算、AI智能 体、大模型等方面进行了深入布局和研发。 ...

Core Viewpoint - The company, Lian Technology, provides comprehensive IDC, cloud computing, and computing power solutions for various mainstream industries, including government, large cloud vendors, internet companies, finance, healthcare, and traditional manufacturing [1] Group 1: Business Offerings - The company offers data center and cloud computing services that support high-performance data center hardware, real-time traffic monitoring, computing power scheduling, and GPU operation and maintenance [1] - The services are designed to cater to large models and institutions, indicating a focus on advanced computing needs [1]

Group 1 - Shanghai Jiyue Xingchen Intelligent Technology Co., Ltd. launched the new generation open-source Agent base model Step3.5Flash, designed for real-time Agent workflow scenarios, balancing inference speed, intelligence level, and cost [1] - Step3.5Flash achieves a maximum inference speed of 350 tokens per second for single request code tasks, providing a "faster, stronger, and more stable" option for Agent base models [1] - The model utilizes a sparse MoE architecture, activating approximately 11 billion parameters per token out of a total of 196 billion parameters, significantly enhancing inference efficiency while maintaining model capability [1] Group 2 - Nearly 10 chip and infrastructure manufacturers, including Huawei Ascend and Alibaba Pingtouge, have completed adaptations for Step3.5Flash, enhancing model adaptability and computing efficiency through collaborative innovation [1] - The establishment of the "MoXin Ecological Innovation Alliance" in July 2025 aims to break down technical barriers between chips, models, and platforms, optimizing performance and accelerating the application of large models across various industries [2] - The industry recognizes that deep collaboration between models and computing power will be a crucial path for the large-scale application of inference models [2]

Core Viewpoint - The release of the new open-source Agent base model Step3.5Flash by Jumpsky is aimed at real-time Agent workflow scenarios, utilizing a sparse MoE architecture with a total parameter count of 196 billion, designed to balance inference speed and cost [1] Group 1: Model Specifications - Step3.5Flash achieves an inference speed of up to 350 tokens per second for single request code tasks [1] - The model activates approximately 11 billion parameters per token [1] Group 2: Industry Collaboration - Multiple chip manufacturers, including Huawei Ascend, Muxi Co., Biran Technology, Suiruan Technology, TianShu Zhixin, and Alibaba Pingtouge, have completed adaptations for the model [1] - Jumpsky previously initiated the "MoCore Ecological Innovation Alliance" in July 2025 with several chip and infrastructure companies to enhance computing efficiency and promote the application of large models [1] Group 3: Strategic Implications - The release of Step3.5Flash is viewed as a further practice in the collaboration between model and computing power [1]