机器之心报道 最近几年，大模型赛道好不热闹。 叫得上名字的几乎数都数不过来：从 GPT、Llama、Gemma、Mistral，到 DeepSeek、Qwen、Kimi、GLM、MiniMax 等等，新模型几乎以周更的速度出现。 但问题是，当架构创新越来越多时，理解它们反而变得越来越困难。不同论文里的模型结构图风格各异、模块命名不统一，即便是研究者，也很难快速看清一个 模型究竟在哪些地方做出了关键改动。 如果把过去几年主流模型的架构放在一起，你会发现一个明显的空白：我们拥有大量模型，却缺少一张清晰的大模型架构图。 最近，AI 研究者 Sebastian Raschka 就尝试给了这样一张图，他将过去几年主流大模型的结构重新绘制，并整理成了一个在线图谱 「LLM Architecture Gallery」。 原文地址：https://sebastianraschka.com/llm-architecture-gallery/ #card -olmo-2-7b 根据 Raschka 介绍，该网站汇集了他此前两篇博客中的内容，这两篇博客分别为《The Big LLM Architecture Comparison》 ...

Summary of Key Points from the Conference Call Industry Overview - **Industry Focus**: The discussion centers on the AI industry, particularly in China and the US, highlighting diverging trends in AI agent adoption and monetization strategies [1][2]. Core Insights - **AI Agent Adoption**: The year 2026 is projected as a pivotal year for the scaled adoption of AI agents, with a shift from chat-based interactions to actionable functionalities. In the US, there is a focus on enterprise adoption, while China is investing heavily in consumer-facing AI products [1]. - **Consumer Engagement**: Major Chinese internet companies launched significant promotional campaigns during the Spring Festival, with Tencent, Alibaba, Baidu, and ByteDance distributing substantial red packets to drive traffic to their AI offerings. For instance, Tencent's Yuanbao distributed Rmb1 billion, Alibaba's Qwen Rmb3 billion, and Baidu's Ernie Rmb500 million [1][6]. - **User Growth**: The campaigns resulted in over 130 million users trying AI services, with notable increases in daily active users (DAU) and monthly active users (MAU) for these platforms [6]. Company-Specific Developments - **Model Launches**: A wave of new AI models was released around the Chinese New Year, with advancements in coding and multimodal capabilities. Notable models include Zhipu's GLM-5.0, Moonshot's Kimi 2.5, and ByteDance's Doubao 2.0, showcasing improved performance and efficiency [3][23]. - **Market Positioning**: Companies like MiniMax are well-positioned to benefit from AI trends, with a focus on full-stack AI capabilities. Baidu and Alibaba are favored for their comprehensive AI ecosystems [4]. Monetization Trends - **Global Enterprise AI Monetization**: There is a surge in AI monetization globally, with companies like Anthropic raising revenue forecasts significantly due to advancements in their AI models. China's AI models are gaining market share in the global enterprise API market, leveraging cost-performance advantages [2]. Competitive Landscape - **AI Disruption Sentiment**: The ongoing narrative of AI disruption is fostering positive sentiment for model providers while making investors cautious about vertical platforms. The competitive landscape is evolving, with intensifying competition and regulatory risks being key concerns for companies like Baidu, Alibaba, Tencent, and Kuaishou [4][24][26][27][29]. Risks and Challenges - **Key Risks**: The report identifies several risks for the internet sector in China, including competition, technology trends, monetization uncertainties, and regulatory changes. Specific risks for individual companies include execution challenges, rising costs, and management issues [24][25][26][27][28][29]. Conclusion - The AI industry in China is experiencing rapid growth and transformation, driven by consumer engagement and technological advancements. Companies are strategically positioning themselves to capitalize on these trends, although they must navigate a complex landscape of competition and regulatory challenges.

Core Insights - The rise of "OpenClaw" in the AI sector has led to a significant increase in demand for tokens, benefiting major model vendors like Zhipu and MiniMax, whose market valuations have soared [1][6][8] - The competition is shifting from merely selling tokens to developing localized tools and intelligent agents that secure user data and business contexts [2][21][31] - The industry is witnessing a transformation from a "cloud-based selling" model to a "localized development" approach, indicating a strategic pivot among major players [3][20] Group 1: Market Dynamics - Major AI companies have experienced unprecedented revenue growth during the Spring Festival, with Zhipu's stock price surging nearly 43% on February 20, 2026, pushing its market cap over 323.2 billion HKD [8][9] - MiniMax also saw a strong market performance, with its valuation surpassing 300 billion HKD, driven by a high percentage of overseas revenue [9][10] - Kimi, a unicorn startup, raised over $700 million in funding, doubling its valuation to over $10 billion, showcasing the rapid growth potential in the AI sector [11] Group 2: Challenges and Risks - The pure token-selling model is facing vulnerabilities due to low technical barriers and price competition, leading to potential market instability [5][31] - OpenClaw and similar frameworks have been criticized for their security risks and unstable performance in real-world applications, raising concerns among enterprise users [17][18][19] - The reliance on open-source frameworks exposes companies to significant security threats, as they require high-level access to sensitive data [18][19] Group 3: Future Trends - The future of AI competition will hinge on the ability to integrate deeply into users' local workflows, moving beyond simple API calls to more sophisticated, localized tools [21][22][28] - Companies are expected to develop proprietary tools that enhance user experience and security, creating a competitive edge in the market [26][27][30] - The market for API services is likely to evolve into a tiered structure, with high-end APIs maintaining premium pricing while simpler tasks are handled locally to reduce costs [32][34]

AI now unveiling its first model to run entirely on chips from the startup Cabus. It's a sign a company's diversifying beyond Nvidia GPUs. Our Dur Debosa has more on that in today's tech check.Morning D. >> Hey, good morning Carl. So never mind that OpenAI is one of Nvidia's largest customers.This is also part of a larger trend. Google shipped Gemini 3 in December trained and served on its own custom AI chips TPUs. Then you got Chinese AI lab GPU releasing GLM trained on Huawei chips and we know that others ...

Core Viewpoint - The article provides a detailed guide on configuring the Clawdbot model on Macmini, highlighting common issues and solutions during the setup process. Configuration Process - The preferred method for configuration is using the command `openclaw configure`, which resolves most issues [6][7]. - During the configuration, users are prompted to select between local or remote setup, choose the model, and input their API Key [9]. Model Selection - There are specific selections for models: for Minimax M2.1, select Minimax; for Kimi K2.5, select moonshot AI [11][12]. - After selecting a model, users should navigate through options using the arrow keys to find the highlighted selection [13]. - Kimi has a coding plan option, while Minimax does not, even if the user has a coding plan membership [14]. Domestic and International Versions - A critical point is the distinction between domestic and international versions of Minimax; users must select 'cn' for domestic coding plan members and the version without 'cn' for overseas members [16]. - Incorrect selections can lead to configuration issues, which can be rectified by manually editing the configuration file [17]. Configuration File Editing - The configuration file is located at `/Users/your_username/.openclaw/openclaw.json`, where users can modify the `baseURL` [18]. - The correct URLs are: domestic version - `api.minimaxi.com`, international version - `api.minimax.io` [23]. Model Switching - After configuration, switching models is straightforward using the command `/model` in the TUI interface, which is initiated with `openclaw tui` [27]. - It is advisable to open a new window with the `/new` command before switching models to avoid issues [29]. Output Issues - The "no output" problem may occur after switching models, indicating that the output is directed to another environment rather than a configuration failure [30]. - Users should check other environments, such as web platforms, to confirm successful configuration [31]. Supported Models - The Clawdbot currently supports three major domestic models, all of which have been successfully configured: Kimi (domestic version), Minimax (international version), and GLM (international version) [34]. Summary of Configuration Steps - The core steps for configuring Clawdbot models are: use `openclaw configure`, manually edit the `baseURL` if necessary, and switch models using the `/model` command [37].

Core Viewpoint - Investment in AI is considered the most certain opportunity for the next twenty years in China, as highlighted by Qiming Venture Partners [3][13]. Group 1: Investment Achievements - Qiming Venture Partners has invested over 12 billion RMB in more than 100 AI projects since 2013, marking a significant presence in the AI sector [5]. - The firm has witnessed a series of IPOs in the AI field, including five notable companies in the past seven months, indicating a solid harvest period after years of strategic investment [3][6]. - Key IPOs include Yunzhisheng, which went public in June 2025, and Wallen Technology, which became the first GPU stock on the Hong Kong Stock Exchange in January 2026 [6][9]. Group 2: Investment Philosophy - Qiming's investment philosophy is encapsulated in the concept of "half a step faster," allowing the firm to identify and invest in key technological and market inflection points before they become mainstream [11][12]. - The firm emphasizes the importance of understanding both the disruptive potential of technology and the timing of market demand to make informed investment decisions [11][12]. Group 3: Future Outlook - Qiming Venture Partners believes that the AI industry in China is still in its early stages, with significant growth potential as applications begin to emerge and costs decrease [13][14]. - The firm is shifting its focus towards application layers in AI, anticipating a diverse range of native applications and super platforms to emerge in various sectors, particularly in healthcare [14].

Core Insights - The event "Praise for China's Economy - Entrepreneur Night" was held on January 29 in Beijing, where Liu Debing, Chairman of Zhipu, received recognition [1][7] - The event highlighted the significant contributions of Zhipu in artificial intelligence (AI) foundational research and key technologies, emphasizing their commitment to independent R&D [3][9] Group 1: Company Achievements - Zhipu has made remarkable efforts in AI foundational research, establishing a strong presence in the global market with their open-source models [3][9] - The company launched its algorithm architecture GLM in 2021, and the release of GLM4.7 this year has positioned their model among the world's leading technologies, laying a solid foundation for pursuing Artificial General Intelligence (AGI) [6][12] Group 2: Leadership Perspectives - Liu Debing expressed that the honor received is a collective achievement of the entire Zhipu team, highlighting their long-term commitment and investment in making machines think like humans [5][12] - Chen Xudong, Chairman of IBM Greater China, acknowledged Zhipu's pioneering role in the market with their large model, wishing them continued success [3][9] - Chen Wei, Chairman of CITIC Publishing Group, emphasized the importance of having outstanding entrepreneurs and tech leaders in the new wave of technology [4][10]

Core Viewpoint - The event "Praise for China's Economy - Entrepreneur Night" was held on January 29 in Beijing, where Liu Debing, Chairman of Zhipu, received recognition for the company's contributions to AI innovation and development [1][5]. Group 1: Company Achievements - Zhipu has made significant efforts in AI foundational research and key technologies, choosing a challenging path of independent research and development rather than simply following trends [3][8]. - The company launched its algorithm architecture GLM in 2021, and the release of GLM4.7 this year has positioned its model among the world's leading technologies, laying a solid foundation for pursuing AGI (Artificial General Intelligence) [6][10]. Group 2: Leadership and Vision - Liu Debing emphasized that the honor received belongs to the entire Zhipu team, highlighting their long-term commitment and dedication to making machines think like humans, which has been the company's vision since its inception [5][10]. - The letter "Z" in Zhipu symbolizes the ultimate goal, representing the aspiration to reach the pinnacle of intelligence in the exploration of AGI [9][10]. Group 3: Industry Recognition - Chen Xudong, Chairman of IBM Greater China, acknowledged Zhipu as the first market-oriented large model in China, expressing hope for the company's future achievements [3][8]. - Chen Wei, Chairman of CITIC Publishing Group, expressed the desire for the presence of outstanding entrepreneurs and tech leaders in the new wave of technology [4][8].

Core Insights - The event "Praise for China's Economy - Entrepreneur Night" was held on January 29 in Beijing, where Liu Debing, Chairman of Zhipu, received recognition [1][5] - The event highlighted the significant contributions of Zhipu in artificial intelligence (AI) foundational research and key technologies, emphasizing their commitment to independent R&D [2][7] Group 1: Company Achievements - Zhipu has made remarkable efforts in AI foundational research, establishing a strong presence in the field [2][7] - The company has chosen a challenging path of independent research and development, moving away from simple follow-the-leader strategies [2][7] - The launch of their algorithm architecture GLM in 2021 and the subsequent release of GLM4.7 have positioned their models among the world's leading technologies, laying a solid foundation for pursuing Artificial General Intelligence (AGI) [4][8] Group 2: Leadership Perspectives - Liu Debing expressed that the honor received belongs to the entire Zhipu team, highlighting the long-term commitment and perseverance of the team in their journey from a Tsinghua University laboratory to the market [4][8] - Chen Xudong, Chairman of IBM Greater China, acknowledged Zhipu's pioneering role in the market with their large model, wishing them continued success [2][7] - Chen Wei, Chairman of CITIC Publishing Group, emphasized the importance of having outstanding entrepreneurs and tech leaders in the new wave of technology [3][7]

Core Viewpoint - The article emphasizes that Qiming Venture Partners has successfully invested in over 100 AI-related companies, with a total investment of approximately 12 billion RMB, marking a significant return on investment as the AI industry matures in China [5][6]. Investment Achievements - Qiming Venture Partners has seen a surge in IPOs in the AI sector, with three IPOs occurring within a short span from late 2025 to early 2026, and a total of five IPOs in the past seven months [3][4]. - The firm has strategically invested across the entire AI value chain, from infrastructure to application layers, demonstrating a comprehensive understanding of the industry's evolution [4][6]. Investment Philosophy - The investment philosophy of Qiming Venture Partners is encapsulated in the concept of "half a step ahead," allowing them to identify and capitalize on key technological and market inflection points before they become widely recognized [12][13]. - This approach requires a deep understanding of both the disruptive potential of technologies and the timing of market demand, enabling the firm to make decisive investments at critical moments [13][14]. Future Outlook - Qiming Venture Partners believes that AI represents the greatest certainty for investment in the next 20 years, with confidence that the Chinese AI industry is not in a bubble but rather at the brink of significant growth [16][18]. - The firm anticipates a shift towards application-layer investments as the AI technology stack matures, predicting a vibrant future for AI applications across various industries, particularly in healthcare [18][19].