二月全球大模型密集迭代，看好AI大模型和应用投资机会【东吴传媒互联网张良卫团队】 智谱GLM-5 Coding Plan套餐上线即售罄，20天内提价30%；MiniMaxM2.5开源一天内全球用户构建超过1万个专 家。 智谱启动算力合伙人招募计划，GLM-5已完成与华为昇腾、摩尔线程、寒武纪、昆仑芯等国产芯片深度适配。 2）Agent能力成为模型迭代核心方向。 近期重要模型及应用事件： 1）2月全球发布17款重磅大模型，中美科技巨头密集迭代。 OpenAI发布GPT-5.3-Codex、Google推出Gemini 3.1 Pro、Anthropic上线Opus 4.6和sonnet 4.6；国内智谱发布GLM-5、 MiniMax上线M2.5、阿里推出Qwen 3.5 二月全球大模型密集迭代，看好AI大模型和应用投资机会【东吴传媒互联网张良卫团队】 近期重要模型及应用事件： 1）2月全球发布17款重磅大模型，中美科技巨头密集迭代。 OpenAI发布GPT-5.3-Codex、Google推出Gemini 3.1 Pro、Anthropic上线Opus 4.6和sonnet 4.6；国内智谱发布GLM-5、 ...

编辑｜杜伟 多年前，苹果用「There's an app for that」开启了移动互联网的黄金时代。从那之后，一个个应用图标统治了我们的数字生活。 而如今，随着 LLM、Agent 的快速发展，这一切正在发生变化。 就在昨天，AI 大神 Karpathy「现身说法」，并抛出了一个激进的观点： 未来的应用不应该是被「下载」的，而应该是被「即兴创作」的 。 他以自己正在进行的有氧运动为例，没有选择去应用商店搜索任何一款「心率管理工具」，而是直接命令 AI 逆向工程了跑步机的云端 API，为自己量身定制了一 个为期八周的、极其私密的实验仪表盘。 这释放出了一个明显的信号：软件的本质正在从现成的商品降维成瞬时的服务。 那么问题来了： 当应用可以随用随建，我们还需要那个臃肿的应用商店吗 ？ | RHR 50 -> 45 Experiment | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 8-week zone 2 + HIIT plan · Feb 18 - Apr 14, 2026 · Dashboard | ...

原文链接：https://x.com/chaowxyz/status/2024358080910413973 声明：本文为转载内容，读者可通过原文链接获得更多信息。如作者对转载形式有任何异议，请联系我 们，我们将按照作者要求进行修改。转载仅用于信息分享，不构成任何投资建议，不代表吴说观点与立 场。 全文如下： --这不仅是合约能力测评，更是 Agent 的链上生存考试。 早上起床收到了一大堆私信，吓得我还以为 AGI 实现了。仔细看是 OpenAI 新发布了智能合约基准测 试，简单讲讲这个。 一句话总结，Agent 理解、修复和运用智能合约的能力，并不是用来抢加密安全公司饭碗的。在我看 来，这些能力指向的是一个更根本的问题：Agent 未来能否真正在加密环境中生存和行动。而 OpenAI 发布的 evmbench，就是衡量这种生存能力的刻度尺。 （来源：吴说） 作者：@chaowxyz Agent 利用加密网络，不只是一种可能性，某种程度上是一种必然。OpenAI 当然也看得到这一点。在 发布报告里也很明确提到 "we expect agentic stablecoin payments to grow" 但 ...

Core Viewpoint - MiniMax, a leading company in general artificial intelligence, has seen its stock price surge, reaching a historical high of 886 HKD, marking a 413% increase since its listing on January 9, 2023, and establishing itself as the top performer in the Hong Kong stock market for companies with a market capitalization over 100 billion HKD [1][4]. Group 1: Stock Performance - MiniMax's stock price hit a peak of 886 HKD during trading, closing at 847 HKD, with a total market capitalization exceeding 265.6 billion HKD [1][4]. - The stock has shown a cumulative increase of over 413% since its debut on the Hong Kong Stock Exchange [1][4]. Group 2: Technological Advancements - The strong stock performance is attributed to significant technological breakthroughs in its flagship model M2.5, particularly in programming and Agent scenarios [1][4]. - M2.5 is positioned as the world's first production-grade model designed natively for Agent scenarios, supporting full-stack programming development across PC, App, and cross-platform applications [1][4]. Group 3: Business Impact - In real business scenarios, M2.5 autonomously completes 30% of overall tasks across various functions, including R&D, product, sales, HR, and finance, with a rising penetration rate [3][7]. - M2.5 has generated 80% of new submitted code in programming scenarios, showcasing its effectiveness [3][7]. Group 4: Product Launches - On February 12, 2023, MiniMax officially launched M2.5, which has achieved industry-leading performance in programming, tool invocation, search, and office applications [1][5]. - A new version, MiniMax-M2.5-highspeed, was launched on February 15, 2023, supporting 100 TPS rapid inference, achieving three times the speed of similar products, attracting significant attention from developers [3][7].

Core Insights - OpenClaw has rapidly gained popularity on GitHub, reaching 190,000 stars, making it the fastest-growing open-source AI project in history, surpassing notable projects like Kubernetes and the Linux kernel [5][8] - OpenClaw represents a significant milestone in the evolution of AI agents, fostering a collaborative ecosystem rather than mere imitation, and igniting entrepreneurial enthusiasm in the AI space [8][9] - The emergence of OpenClaw is reshaping the investment landscape, with projects related to it experiencing rapid valuation increases as investor interest surges [8][9] Interaction Paradigm Shift - OpenClaw offers a qualitative leap in user interaction compared to traditional AI assistants like ChatGPT, embedding itself within existing communication tools and operating seamlessly in the background [11][12] - This shift transforms AI from a one-time tool into a long-term partner, emphasizing user control and ownership over AI as a personal digital asset rather than a service provided by centralized platforms [12][13] Revaluation of AI Models and Agents - OpenClaw's significance lies in its ability to detach agents from platforms, allowing for personal ownership and expansion, which challenges the prevailing narrative that large models are the sole value drivers in AI [15][16] - The platform enables users to freely combine models and tools, shifting power from centralized companies to developers and users, thus redefining the valuation of both agents and models in the industry [15][16] New Entrepreneurial Opportunities - The rise of OpenClaw is creating new entrepreneurial avenues, including security systems, AI NAS, and independent agent memory solutions, as the demand for personal AI capabilities grows [20][21][22] - The concept of treating AI as a "digital life" opens up various sectors, such as security, infrastructure, and memory management, which are becoming essential for the next generation of AI agents [20][22]

Core Insights - The AI industry is undergoing a transformation where the focus has shifted from static performance metrics to the ability to rapidly evolve and adapt, redefining competitive advantages [2][24][25] - MiniMax M2.5 exemplifies this trend by achieving high performance at a significantly reduced cost, indicating a new paradigm in AI model development [3][23] Group 1: Evolution of AI Standards - The emergence of MiniMax M2.5 highlights a new competitive landscape where the speed of evolution is the key variable for success, rather than just current performance [17][24] - The AI competition is transitioning from a pre-training phase focused on knowledge accumulation to a post-training phase centered on practical execution and problem-solving [9][10] Group 2: Performance Metrics - MiniMax M2.5 achieved an 80.2% score on the SWE-Bench Verified benchmark, closely rivaling the top competitor Claude Opus 4.6, which scored 80.8% [3][11] - The model operates at a cost of only $1 per hour for continuous operation at 100 TPS, making it significantly cheaper than its peers [6][23] Group 3: Technological Advancements - The rapid evolution of M2.5 is evident, with scores improving from 74.0% in M2.1 to 80.2% in M2.5 over a span of 108 days [19][20] - MiniMax's Forge system is designed to accelerate the evolution of AI models, allowing for efficient adaptation to various real-world environments [21][22] Group 4: Business Implications - The low cost and high efficiency of M2.5 are reshaping the cost-benefit model for AI applications, making AI a viable labor force alternative [23] - The introduction of M2.5 signals a shift in the industry’s focus from static performance to dynamic evolution capabilities, emphasizing the importance of a robust evolutionary system [24][25]

Core Insights - AI application concept stocks experienced fluctuations, with Zhangyue Technology hitting a limit-up for the fifth consecutive day, while other stocks like Guotou Intelligent and Wangsu Technology also saw gains [1] - Domestic AI companies are witnessing a surge in model releases, exemplified by MiniMax's announcement of its new text model, MiniMax M2.5, which allows for cost-effective operation of complex agents [1][2] Group 1: Market Trends - The cloud computing ETF Huaxia (516630) and the entrepreneurial board AI ETF Huaxia (159381) are undergoing adjustments due to the pullback in the optical module CPO and data center sectors [1] - The entrepreneurial board AI ETF Huaxia has attracted over 500 million yuan in the last 60 days, indicating strong investor interest [1] Group 2: AI Model Developments - MiniMax's M2.5 model can operate complex agents at a cost of only $0.3 per hour, suggesting that $10,000 can sustain four agents for an entire year [1] - The primary challenge for the M2 series models is the pace of capability improvement, highlighting the competitive landscape in AI model development [1] Group 3: Investment Opportunities - Huatai Securities reports that the acceleration of Agent applications is expected to peak in 2026, driven by rapid iterations in Agentic Coding and competition among major companies for personal agent assistants [2] - The entrepreneurial board AI ETF Huaxia has a high elasticity with nearly 50% weight in CPO, covering domestic software and AI application companies [2] - The cloud computing ETF Huaxia focuses on domestic AI software and hardware, with a combined weight of 83.7% in computer software, cloud services, and computer equipment [2]

Summary of Conference Call Notes Industry Overview - The conference call discusses the cloud computing industry, highlighting recent price increases by major players such as AWS and Google Cloud, as well as domestic companies like UCloud [1][2]. Key Points and Arguments - Recent price hikes by AWS and Google Cloud reflect rising upstream hardware costs and confirm strong downstream demand [3]. - The demand for Agents and multimodal applications is driving a surge in Token consumption, particularly with the popularity of OpenClaw (Clawdbot) [3]. - The proliferation of multimodal AI applications, such as Seedance 2.0, is expected to further accelerate Token consumption, leading to increased usage of cloud computing resources and guaranteed revenue for cloud providers [3]. Important Companies Mentioned - **Kingsoft Cloud**: Recommended as a core cloud provider within the Xiaomi-Kingsoft ecosystem, noted for its high elasticity [4]. - **Baidu Group**: Recognized for its full-stack AI capabilities and control over Kunlun Chip [4]. - **Alibaba**: Identified as a leading domestic cloud provider with strong technical capabilities in full-stack AI and significant cost advantages from self-developed chips [4]. - **Tencent Holdings**: Mentioned for achieving profitability in Tencent Cloud after 25 years, with differentiated competition and SaaS products contributing to 25% of revenue [4].

Summary of Conference Call Notes Industry Overview - The focus is on the global large model manufacturers, particularly in the fields of Coding and Agent capabilities, indicating a significant surge in demand and development within this sector [1] Core Insights and Arguments - The explosion of Agent technology is confirmed as a major trend for the year, with an inevitable increase in both token consumption and pricing [1] - The concept of "computing power inflation" is highlighted as a central theme for the year, suggesting that the demand for computational resources will continue to rise [1] - Continuous non-linear growth in token consumption is anticipated, indicating a robust market outlook for this segment [1] Key Components of Computing Power Inflation - The following areas are identified as critical components contributing to computing power inflation: - GPU (Graphics Processing Unit) - Storage - CPU (Central Processing Unit) - Networking - AI Infrastructure (notable companies include Wangsu Science & Technology, Deepin Technology, Yuke Technology, Kingsoft Cloud, Capital Online, and Qingyun Technology) [1] Model Manufacturers - Key players in the model manufacturing space are mentioned, including: - Zhipu AI - Minimax - iFlytek [1] This summary encapsulates the essential points from the conference call, focusing on the industry dynamics, core insights, and significant players involved in the large model manufacturing sector.

在以每秒输出100个token的情况下，连续工作一小时只需要1美金，而在每秒输出50个 token的情况下， 只需要0.3美金。也就是说，1万美金可以让4个Agent连续工作一年。M2.5提供了经济上几乎无限制地构 建和运营Agent的可能性。对于M2系列的模型来说，唯一的问题变成了模型能力的进步速度。 （文章来源：第一财经） MiniMax微信公众号消息，MiniMax介绍，设计M2系列的初衷是希望能没有成本约束的情况下运行复杂 Agents，随着能力的持续提升，M2.5已经接近实现了这个目标。M2.5提供两个效果一样，但是速度和 价格不一样的版本：速度100TPS左右的快速版本，处理每百万的token输入只需要0.3美金，处理每百万 的token输出只需要2.4美金。而50TPS 的版本的输出价格还更低一倍。按照输出价格参考，50 TPS的版 本价格是Opus、Gemini 3 Pro以及GPT5这些模型的1/10-1/20。 ...