Core Insights - Zhiyu Technology has officially listed on the Hong Kong Stock Exchange, becoming the first publicly traded company focused on general AI models, marking a significant milestone for the industry [3][5]. - The IPO was highly successful, with a subscription rate of 1159.46 times for the public offering and 15.28 times for the international offering, raising over HKD 4.3 billion [3][10]. - The company aims to develop general AI models that enable machines to think like humans, a vision articulated by its chairman, Liu Debing [3][5]. Company Overview - Founded in June 2019, Zhiyu Technology has grown to become the largest independent developer of general AI models in China, ranking second among all general AI model vendors, including major internet companies [5][8]. - The company has a strong backing from notable investors, including Meituan, Ant Group, Tencent, and Lenovo, which collectively hold shares through various entities [8][10]. Financial Performance - Zhiyu reported revenues of RMB 0.57 million in 2022, RMB 1.25 million in 2023, and projected revenues of RMB 3.12 million for 2024, with a significant net loss of RMB 62.47 billion over three and a half years [11][13]. - R&D expenses have escalated, reaching RMB 21.95 million in 2024, indicating a focus on high investment and heavy R&D [11][13]. - The company has accumulated a total revenue of RMB 6.85 million and a net loss of RMB 62.47 billion since its inception [13]. Business Model and Revenue Structure - Zhiyu's revenue is primarily derived from localized deployments, which accounted for over 84% of total revenue in the past three and a half years, indicating a strong presence in government and large enterprise sectors [14][15]. - The company offers both localized and cloud deployment options for its models, with revenue recognition occurring upon delivery and acceptance by clients for localized deployments [25][27]. Market Position and Competitive Landscape - Zhiyu Technology is positioned against major competitors such as iFlytek, Alibaba, SenseTime, and Baidu, with a focus on developing a comprehensive AI ecosystem [5][6]. - The company has established a significant technological barrier with its proprietary GLM architecture, which has been recognized in international academic publications [22][23]. Research and Development - The R&D team consists of over 650 members, representing 74% of the total workforce, with a strong academic background from top institutions like Tsinghua University [20][21]. - Zhiyu has published approximately 500 high-impact papers, with over 58,000 citations, and holds 86 authorized patents, showcasing its commitment to innovation [22][23]. Customer Base and Token Consumption - As of 2025, Zhiyu serves over 12,000 institutional clients, with a daily token consumption of 4.2 trillion, reflecting its growing market adoption [35][37]. - The concentration of revenue from the top five clients has decreased over the years, indicating a diversification of its customer base [37][39].

闻乐 发自 凹非寺 量子位 | 公众号 QbitAI 刚刚，全球大模型第一股，终于在港交所敲钟！ 被称为 中国版OpenAI 的 智谱 正式挂牌上市（股票代码2513）。 不仅拿下 全球首家AGI基座模型上市公司 头衔，首日开盘 涨超3% ，报 120港元/股 ，市 值突破 528亿港元 。 | HK 02513 智谱 | | | | | | | | --- | --- | --- | --- | --- | --- | --- | | 120.000港元 +3.800 +3.27% | | | | | | | | ○ 集合竞价 01-08 09:22:00 北京时间 | | | | | | | | 今开 最高 | 120.000 | | | 120.000 | | 成交量 612.45万股 | | 昨收 最低 | 116.200 | | | 120.000 | 成交额 | 8.13亿 | | 换手率 | 2.84% | | 市盈(TTM) | | 总市值 | 528.28亿 | | 分时 | 五日 | 日K | | 月K | 零K 2 每次号:重要位 | | 这波2026年的重磅开场，直接把AI资本市场的热 ...

Core Viewpoint - The company Zhiyu, known as the Chinese version of OpenAI, has officially listed on the Hong Kong Stock Exchange, becoming the world's first publicly traded AGI foundational model company, with a market capitalization exceeding 52.8 billion HKD on its first trading day [1][21]. Group 1: IPO Details - Zhiyu's IPO raised over 4.3 billion HKD, with an oversubscription rate of 1159.46 times for the public offering and 15.28 times for the international offering [4][25]. - The stock opened at 120 HKD, up over 3% from the previous close of 116.2 HKD, with a trading volume of 6.1245 million shares [2][22]. - The company attracted 11 cornerstone investors, including major state-owned enterprises and top insurance firms, securing 29.8 billion HKD in subscriptions, which accounted for nearly 70% of the offering [9][28]. Group 2: Technological Strength and Market Position - Zhiyu's flagship model, GLM-4.7, has achieved top rankings in various global AI indices, outperforming competitors like GPT-5.2 [29]. - The GLM architecture is compatible with over 40 domestic chip types, and its AutoGLM 2.0 can control 80 million devices, with an average daily token usage of 4.6 trillion [31]. - The company has seen its revenue grow significantly, with a compound annual growth rate of 130%, increasing from 57.4 million CNY in 2022 to 312.4 million CNY in 2024 [33]. Group 3: Investment and R&D Focus - Zhiyu has invested over 4.4 billion CNY in research and development from 2022 to mid-2025, with 74% of its workforce dedicated to R&D [37]. - The company plans to allocate 70% of its IPO proceeds to continue R&D in large models and 10% to optimize its MaaS platform [39]. - The business model has attracted over 2.7 million enterprises and developers in China, with nine of the top ten internet companies utilizing its platform [36].

Core Viewpoint - Zhipu AI, known as the "Chinese version of OpenAI," is set to become the world's first publicly listed large model company, with its IPO scheduled for January 8, 2026, on the Hong Kong Stock Exchange under the stock code 2513 [7][8]. Group 1: IPO Details - Zhipu AI has officially launched its IPO, aiming to raise approximately HKD 4.3 billion, with a post-listing market valuation expected to exceed HKD 51.1 billion [3][11]. - The IPO will issue a total of 37,419,500 H-shares, with 1,871,000 shares available for public sale in Hong Kong and 35,548,500 shares for international sale [6][10]. - The offering price is set at HKD 116.20 per share, with the subscription period running from December 30, 2025, to January 5, 2026 [9][11]. Group 2: Financial Performance - Zhipu AI has achieved significant revenue growth, with revenues of RMB 57.4 million, RMB 124.5 million, and RMB 312.4 million from 2022 to 2024, representing a compound annual growth rate of 130% [27]. - The company reported a revenue of RMB 191 million in the first half of 2025, marking a year-on-year increase of 325% [27]. - Zhipu AI has maintained a gross margin above 50% over the past three years, outperforming the industry average of approximately 40% [31][32]. Group 3: R&D Investment - The company has heavily invested in research and development, with R&D expenses rising to RMB 844 million, RMB 5.289 billion, and RMB 21.954 billion from 2022 to 2025, indicating a significant commitment to innovation [35]. - At its peak, R&D spending reached eight times the company's revenue for the period, highlighting the financial pressures associated with high R&D costs in the AI sector [36]. Group 4: Market Position and Strategy - Zhipu AI has established a strong market presence, with over 12,000 enterprise clients and more than 80 million end-user devices powered by its models [26]. - The company has successfully implemented a MaaS (Model as a Service) business model, which has attracted over 270,000 enterprises and application developers in China, with nine of the top ten internet companies utilizing its models [25][26]. - The latest flagship model, GLM-4.7, has achieved top rankings in various AI performance evaluations, further solidifying Zhipu's position in the competitive landscape [18][19]. Group 5: Founding and Leadership - Zhipu AI was founded in 2019, originating from Tsinghua University's technology transfer, with a leadership team composed of experts from the university's Knowledge Engineering Laboratory [41][53]. - The CEO, Zhang Peng, and Chief Scientist, Tang Jie, are both prominent figures in the AI research community, contributing to the company's technological advancements and strategic direction [46][51].

文 | AIX财经（AIXcaijing），作者 | 金玙璠，编辑 | 魏佳 一纸招股书，揭开了"中国版OpenAI"智谱AI的神秘面纱，也把一个备受关注的话题抛到台前：在中国 大模型的牌局中，智谱AI究竟排第几？ 12月19日，这家顶着"清华系"光环的明星公司在通过港交所聆讯后，披露了招股书，向"全球大模型第 一股"发起最后冲刺。 一时间，所有人的目光都聚焦在这家成立仅6年的年轻公司身上，最受关注的，除了它豪华的股东阵容 （美团、阿里、腾讯、小米）和超过240亿元的估值，还有一张颇具话题性的中国大模型厂商"座次 图"。 排名背后，另一组数字同样醒目。2024年，智谱AI经调整净亏损达24.7亿元。2025年上半年亏损仍在加 剧，营收1.9亿元对应经调整净亏损17.5亿元。 高速扩张与持续亏损并行，几乎是当前大模型公司的共同特征，也折射出这一行业高研发、长周期的普 遍挑战。 因此，要看清智谱AI的真实位置，我们或许不能只依赖一份报告中的单一维度排名，而需要回到三个 更本质的问题：智谱AI究竟是怎样一家公司？它的财务状况说明了什么？以及，牌桌之上真实的战局 究竟如何？ 智谱AI，是一家怎样的公司？ 要理解智谱A ...

智通财经记者 | 宋佳楠 12月9日，智通财经获悉，"大模型六小虎"之一的智谱AI宣布开源核心AI Agent模型AutoGLM，称希望 通过这款模型，让AI真正学会"使用手机"。 谈及在当下这个时间节点选择开源的原因，智谱表示，"这件事（AI手机）只在一家公司做，是不够 的。"智谱认为，AI手机已经是趋势，但如果"会用手机的AI能力"只掌握在极少数厂商手里，那便意味 着两件事：一是开发者的创新空间，要看平台愿不愿意开放接口；二是用户日常生活中最重要的那台设 备，越来越像"别人家的入口"。 去年10月，智谱发布了首个可在真机上完成一条完整操作链路的AutoGLM。一个月后，AutoGLM发出 了由AI完成的手机红包。 进入2025年，该公司又发布了AutoGLM 2.0，并把它放进一台和用户真实世界脱离的虚拟手机里。基于 此，这台手机跑在云端，每一个动作都可回放、审计、干预，也可以对敏感数据加以隔离。 智谱特别提到，"让AI会用手机之前，我们要先保证，它不会在不该点的地方胡乱伸手。"为此，其主动 放弃了操作微信等这些离用户隐私较近的APP。 但智谱强调，Phone Use能力一旦做大，天然会碰到敏感信息，包括 ...

Core Insights - The core focus of the news is the announcement by Zhipu that it has open-sourced its core AI Agent model, AutoGLM, which aims to enable AI to perform mobile operations effectively, marking a significant advancement in AI capabilities [1][2]. Group 1: AutoGLM Development and Features - AutoGLM is designed to help users perform tasks on mobile applications, such as placing orders on food delivery apps and managing notifications [1]. - The model is set to be released on October 25, 2024, and is recognized as the first AI Agent with "Phone Use" capabilities [1]. - In 2025, Zhipu plans to launch AutoGLM 2.0, which will utilize algorithms like MobileRL, ComputerRL, and AgentRL to enhance the model's accuracy and generalization by conducting reinforcement learning across thousands of virtual device environments [1]. Group 2: Industry Context and Privacy Concerns - Zhipu is a leading AI large model company incubated by Tsinghua University, with a focus on developing next-generation cognitive intelligence models [2]. - The industry is shifting towards more practical AI agents that can not only communicate with users but also take actionable steps based on user input [2]. - Privacy and network security concerns are significant, particularly regarding the need for user authorization for screen reading permissions, which raises fears about personal information leakage and potential misuse by malicious entities [2][3]. Group 3: Competitive Landscape and Challenges - Recently, Doubao announced a technical preview of its mobile assistant, which has garnered market attention and led to a surge in ZTE's stock price [2]. - Users reported issues with the Doubao mobile assistant, including app crashes and login failures, which may be linked to the need for system-level permissions [3]. - Doubao has stated that while its assistant requires screen reading capabilities, it does not store user screen content in the cloud, aiming to ensure user privacy [3]. - Zhipu has opted for a cloud-based deployment for AutoGLM to allow for action replay, auditing, and intervention, thereby isolating sensitive data [3].

- DeepSeek-V3.1 model is an upgraded version of the DeepSeek language model, featuring a hybrid inference architecture that supports both "thinking mode" and "non-thinking mode" for different task complexities[12][13][14] - The model's construction involves dynamic activation of different attention heads and the use of chain-of-thought compression training to reduce redundant token output during inference[13] - The context window length has been expanded from 64K to 128K, allowing the model to handle longer documents and complex dialogues[15] - The model's performance in various benchmarks shows significant improvements, such as a 71.2 score in xbench-DeepSearch and 93.4 in SimpleQA[17] - The model's evaluation highlights its advancements in hybrid inference, long-context processing, and tool usage, although it still faces challenges in complex reasoning tasks[21] - Seed-OSS model by ByteDance features 36 billion parameters and a native 512K long-context window, emphasizing research friendliness and commercial practicality[22][23] - The model uses a dense architecture with 64 layers and integrates grouped-query attention (GQA) and rotary position encoding (RoPE) to balance computational efficiency and inference accuracy[23] - The "thinking budget" mechanism allows dynamic control of inference depth, achieving high scores in various benchmarks like 91.7% accuracy in AIME24 math competition[24] - The model's evaluation notes its strong performance in long-context and reasoning tasks, though its large parameter size poses challenges for edge device deployment[25] - WebWatcher by Alibaba is a multimodal research agent capable of synchronously parsing image and text information and autonomously using various toolchains for multi-step tasks[26][27] - The model's construction involves a four-stage training framework, including data synthesis and reinforcement learning to optimize long-term reasoning capabilities[27] - WebWatcher excels in benchmarks like BrowseComp-VL and MMSearch, achieving scores of 13.6% and 55.3% respectively, surpassing top closed-source models like GPT-4o[28] - The model's evaluation highlights its breakthrough in multimodal AI research, enabling complex task handling and pushing the boundaries of open-source AI capabilities[29] - AutoGLM 2.0 by Zhipu AI is the first mobile general-purpose agent, utilizing a cloud-based architecture to decouple task execution from local device capabilities[32][33] - The model employs GLM-4.5 and GLM-4.5V for task planning and visual execution, using an asynchronous reinforcement learning framework for end-to-end task completion[34] - AutoGLM 2.0 demonstrates high efficiency in various tasks, such as achieving a 75.8% success rate in AndroidWorld and 87.7% in WebVoyager[35] - The model's evaluation notes its significant advancements in mobile agent technology, though it still requires optimization for cross-application stability and scenario generalization[37] - WeChat-YATT by Tencent is a large model training library designed to address scalability and efficiency bottlenecks in multimodal and reinforcement learning tasks[39][40] - The library introduces parallel controller mechanisms and partial colocation strategies to enhance system scalability and resource utilization[40][42] - WeChat-YATT shows a 60% reduction in overall training time compared to the VeRL framework, with each training stage being over 50% faster[45] - The model's evaluation highlights its effectiveness in large-scale RLHF tasks and its potential to drive innovation in multimodal and reinforcement learning fields[46] - Qwen-Image-Edit by Alibaba's Tongyi Qianwen team is an image editing model that integrates dual encoding mechanisms and multimodal diffusion Transformer architecture for semantic and appearance editing[47][48] - The model's construction involves dual-path input design and chain editing mechanisms to maintain high visual fidelity and iterative interaction capabilities[48][49] - Qwen-Image-Edit achieves SOTA scores in multiple benchmarks, with comprehensive scores of 7.56 and 7.52 in English and Chinese scenarios respectively[50] - The model's evaluation notes its transformative impact on design workflows, enabling automated handling of rule-based editing tasks and lowering the barrier for visual creation[52] Model Backtest Results - DeepSeek-V3.1: Browsecomp 30.0, Browsecomp_zh 49.2, HLE 29.8, xbench-DeepSearch 71.2, Frames 83.7, SimpleQA 93.4, Seal0 42.6[17] - Seed-OSS: AIME24 math competition 91.7%, LiveCodeBench v6 67.4, RULER (128K) 94.6, MATH task 81.7[24] - WebWatcher: BrowseComp-VL 13.6%, MMSearch 55.3%, Humanity's Last Exam-VL 13.6%[28] - AutoGLM 2.0: AndroidWorld 75.8%, WebVoyager 87.7%[35] - Qwen-Image-Edit: English scenario 7.56, Chinese scenario 7.52[50]

Investment Rating - The report maintains a "Recommended" investment rating for the semiconductor industry, indicating that the industry index is expected to outperform the benchmark index by more than 10% [7]. Core Insights - The overseas AI chip index fell by 2.23% this week due to the announcement of upcoming semiconductor tariff policies by the U.S. government, leading to declines in major chip stocks such as AMD and Marvell [1][10]. - In contrast, the domestic AI chip index surged by 18.9% following the release of DeepSeek-V3.1, which is tailored for next-generation domestic chip architectures, benefiting leading domestic chip companies [1][10]. - The report highlights significant fluctuations in various semiconductor indices, with the storage chip index rising by 9.6% and the power semiconductor index increasing by 5.7% [1][15]. Market Indices Summary - The overseas AI chip index experienced a decline of 2.23%, with AMD down 5.5% and Marvell down 4.2% [1][10]. - The domestic AI chip index saw an increase of 18.9%, with notable gains from companies like 中芯国际 (10.1%) and 寒武纪 (34.6%) [1][10]. - The server ODM index fell by 5.3%, with Quanta being the only stock to rise by 15.2% [1][10]. - The storage chip index rose by 9.6%, with 兆易创新 increasing by 22.8% [1][15]. - The power semiconductor index increased by 5.7%, with 华润微 rising by 8.3% [1][15]. Industry Data Summary - Taiwan's top four foundries are expected to generate a combined revenue of $35.15 billion in Q3 2025, reflecting a 7.1% quarter-over-quarter growth, but projected to decline to $32.1 billion in Q4 2025, a decrease of 8.7% [2][25]. - The domestic XR consumer market saw sales of 261,000 units in the first half of 2025, a 9% increase quarter-over-quarter but a 21% year-over-year decline, with AR devices showing a 35% year-over-year growth [2][26]. - Global smart glasses shipments surged by 110% year-over-year in the first half of 2025, with Meta holding a 73% market share [2][28][31]. Major Events Summary - The U.S. government announced a $11 billion investment to acquire a 9.9% stake in Intel, becoming its largest shareholder, which includes funds from the CHIPS and Science Act [3][32]. - DeepSeek launched its new AI model, DeepSeek-V3.1, specifically designed for domestic chips, marking a significant step in China's AI industry [3][33]. - Vivo released its first mixed reality headset, Vision Exploration Edition, aimed at everyday use [3][37].

Investment Rating - The report maintains a "Buy" rating for the industry, indicating a positive outlook for the sector's performance [5][49]. Core Insights - The media industry saw a weekly increase of 5.17%, outperforming the Shanghai Composite Index and the CSI 300 Index, which rose by 3.49% and 4.18% respectively [11][19]. - Key companies such as KuaLing AI and Kunlun Wanwei reported significant revenue growth, with KuaLing achieving over 250 million in revenue for Q2 2025, exceeding expectations [2][46]. - The gaming market in China reached a size of 29.084 billion yuan in July 2025, with mobile gaming contributing significantly to this growth [3][25]. - The report highlights the successful release of AI applications and the cultural export theme as key investment themes, particularly in gaming, IP, short dramas, and publishing [4][47]. Market Performance - The media industry ranked 6th among all sectors with a weekly increase of 5.17%, while the gaming sector saw a rise of 6.09% [11][19]. - Notable performers included Guomai Culture and Shunwang Technology, with weekly increases of 24.79% and 24.16% respectively [19][20]. Key Data and Dynamics AI Applications - Recent downloads for AI applications on iOS showed varied performance, with Doubao leading at approximately 209.57 thousand downloads, while DeepSeek experienced a decline of 8.88% [2][23]. Gaming Sector - The mobile gaming market in July 2025 was valued at 21.36 billion yuan, with a year-on-year growth of 0.92% [3][25]. - The overseas revenue from self-developed games reached 1.693 billion USD, marking a year-on-year increase of 6.76% [28][29]. Film Industry - The total box office for the week of August 15-21 was 1.252 billion yuan, with "Wang Wang Mountain Little Monster" leading the box office [41][43]. Investment Recommendations - The report suggests focusing on AI applications and cultural export themes, with specific attention to companies like Giant Network, KuaLing, and Meitu [4][47].