随着大模型能力持续迭代以及AI应用加速落地，脉脉数据显示，2026年以来新发AI岗位量同比增长14 倍。算法工程师、大模型算法工程师、后端开发岗位位列热招的技术岗位前三；非技术岗位则主要集中 在产品、运营与设计方向。 登上央视春晚的多家企业都在脉脉招聘人才，岗位类型覆盖研发、产品、运营等方向，多个岗位年薪超 100万元。宇树科技持续在脉脉社区发布帖子招聘人才。今年春节期间，宇树科技HR发帖称"宇树一直 在招聘研发，现在投递私聊，年后优先处理"。字节跳动的豆包业务给"大模型应用架构专家"开出了128 万元的年薪，"平台产品经理"月薪最高也达到6万元。 另外两家登上春晚的企业松延动力和银河通用，也在脉脉上发布多个高薪岗位，其中"电机控制算法工 程师"、"多模态导航算法工程师"岗位年薪均超过100万元。除AI外，互联网大厂、游戏、智能驾驶行业 的招聘力度也持续增加。 责任编辑：宋雅芳 随着大模型能力持续迭代以及AI应用加速落地，脉脉数据显示，2026年以来新发AI岗位量同比增长14 倍。算法工程师、大模型算法工程师、后端开发岗位位列热招的技术岗位前三；非技术岗位则主要集中 在产品、运营与设计方向。 登上央视春晚的多 ...

春节假期，记者探访上海人工智能企业、北京中试平台、湖北武汉机器人体验店，从不同产业环 节，看蓬勃发展的机器人产业图景如何展开。 上海徐汇区"模速空间"创新社区，无问芯穹智能科技公司大屏上，数据实时跳动，显示着算力运行 状况。 "我们致力于让大模型服务更便捷。"公司高级系统架构师夏磊解释，团队的工作是让不同的大模型 算法高效运行在各种芯片上。 "我们能调用相隔2000公里的机房联合训练，让算力利用率超过97%。"夏磊说，这相当于搭建起高 效、通用的"智力"供给网，为机器人提供"思考""决策"的基础。 人工智能的"大脑"愈发聪慧，机械的"身躯"越来越灵活，不同型号的产品走进店铺、工厂、舞 台……机器人正加速进入产业化的快车道。 中试，是成果转化"从1到10"的关键一跃。没有它，实验室的样品很难走向"从10到100"的大规模生 产。过去，研发机构自建试验线成本高、效率低。如今，这个五层楼的中试验证平台拥有完整生产测试 线，平均每17小时就能组装一台机器人，具备年产5000台的能力。 "中试验证平台破解了量产难题，搭建起从技术研发到产品转化的关键桥梁。"王博说。 湖北武汉光谷，"人形机器人7S店"春节假期不打烊。 店 ...

| 代码 | 名称 | 两日图 | 现价 | 涨跌 | 涨跌幅 | | --- | --- | --- | --- | --- | --- | | 000001 | 上证指数 | 5 | 4139.49 | 22.08 | 0.54% | | 399001 | 深证成指 | 1 | 14361.70 | 70.14 | 0.49% | | 399006 | 创业板指 | VW | 3309.77 | 1.51 | 0.05% | | 000300 | 沪深300 | VW | 4725.91 | 18.36 | 0.39% | | 000016 | FJESO | m | 3054.78 | 13.46 | 0.44% | | 000680 | 科创综指 | W | 1791.83 | -6.38 | -0.35% | | 000688 | 科创20 | | 1457.04 | -8.32 | -0.57% | | 899050 | - REFF 0 | | 1531.61 | -3.83 | -0.25% | | 868008 | 万得微盘股指数 ~~ | | 631109.18 | 5154.25 | 0 ...

Domestic Market Overview - The A-share market opened positively in the Year of the Horse, with the Shanghai Composite Index closing at 4117.41 points, up 0.87%, and the Shenzhen Component Index closing at 14291.57 points, up 1.36% [1][10] - A total of 24 out of 30 sectors in the CITIC first-level industry rose, with the highest gains in oil and petrochemicals, building materials, and non-ferrous metals, while consumer services, media, and computers saw significant declines [1][10] - The total trading volume of the entire A-share market reached 22,182 billion, showing an increase compared to the previous day [1][10] Industry Insights - The report highlights the significant increase in domestic tourism during the 2026 Spring Festival, with 596 million domestic trips taken, an increase of 95 million compared to the previous year, and total spending reaching 803.48 billion, up 126.48 billion [12][13] - The loan market quotation rate (LPR) remained stable for the ninth consecutive month, with the one-year LPR at 3% and the five-year LPR at 3.5% [15][16] - The average daily usage of large models by Chinese enterprises reached 37 trillion tokens, a 263% increase from the previous half-year, with Alibaba Cloud's Qwen model leading with a market share of 32.1% [17]

每经AI快讯，2月25日，港股大模型概念股短线跳水，智谱、MINIMAX-WP盘中均跌超10%。 ...

Core Viewpoint - Cathay Securities maintains a "Buy" rating for MINIMAX-WP (00100) with a target price of HKD 1012, slightly adjusting the company's revenue projections for 2025-2027 to USD 0.7/2.2/5.5 billion, considering its unique position as a global multimodal model in the Hong Kong stock market and the strong catalyst from the recent M2.5 flagship model release, assigning a 186x PS for 2026 [1] Group 1: Technical Capabilities - MiniMax M2.5 leverages MoE architecture optimization and the native Agent RL framework Forge, achieving industry-leading performance in programming, tool invocation, search, and office productivity scenarios, with core metrics comparable to top overseas models like Claude Opus 4.5 [1] - The model's inference speed reaches 100 TPS, which is double the industry average, and SWE-Bench Verified task processing speed has improved by 37% compared to M2.1, demonstrating a dual breakthrough in performance enhancement and efficiency optimization [1] Group 2: Cost Control and Commercial Viability - M2.5 establishes a leading cost advantage through token consumption optimization, parallel tool invocation upgrades, and tiered pricing design, overcoming economic barriers for large-scale Agent operations [2] - The pricing strategy includes a dual-version approach, with a 100 TPS fast version costing only USD 1 for continuous operation for one hour, and a 50 TPS version priced as low as USD 0.3, making it 1/10 to 1/20 the cost of overseas models like Claude Opus, Gemini 3 Pro, and GPT-5, positioning it as a global benchmark for high cost-performance models [2] Group 3: Product Iteration and Ecosystem Development - MiniMax demonstrates industry-leading model iteration speed, completing updates for M2, M2.1, and M2.5 within 108 days, significantly outpacing overseas giants like Anthropic, OpenAI, and Google, thanks to deep synergy between large-scale reinforcement learning and engineering capabilities [3] - The M2.5 model is fully integrated into the MiniMax Agent platform, refining core Office Skills capabilities and enabling users to create reusable industry experts, with over 10,000 custom experts now available covering high-frequency scenarios in office, finance, and programming [3]

Core Viewpoint - Cathay Securities maintains an "overweight" rating for MINIMAX-WP (00100) with a target price of 1012 HKD, slightly adjusting the company's revenue forecast for 2025-2027 to 0.7/2.2/5.5 billion USD, considering its unique position as a global multimodal large model in the Hong Kong stock market and the strong catalyst from the recent M2.5 flagship model release, assigning a 186x PS for 2026 [1] Group 1: Technical Capabilities - MiniMax M2.5 leverages MoE architecture optimization and the native Agent RL framework Forge, achieving industry-leading performance in programming, tool invocation, search, and office productivity scenarios, with core metrics comparable to top overseas models like Claude Opus 4.5 [1] - The model's inference speed reaches 100 TPS, which is double the industry average, and SWE-Bench Verified task processing speed has improved by 37% compared to M2.1, demonstrating a dual breakthrough in performance enhancement and efficiency optimization [1] Group 2: Cost Control and Commercial Viability - M2.5 establishes a leading cost advantage through token consumption optimization, parallel tool invocation upgrades, and tiered pricing design, overcoming economic barriers for large-scale Agent operations [2] - The pricing strategy includes a dual-version approach, with a 100 TPS fast version costing only 1 USD for continuous operation for one hour, and a 50 TPS version priced as low as 0.3 USD, making it 1/10 to 1/20 the cost of overseas models like Claude Opus, Gemini 3 Pro, and GPT-5, positioning it as a global benchmark for high cost-performance large models [2] Group 3: Product Iteration and Ecosystem Development - MiniMax demonstrates industry-leading model iteration speed, completing updates for M2, M2.1, and M2.5 within 108 days, significantly outpacing overseas giants like Anthropic, OpenAI, and Google, thanks to deep synergy between large-scale reinforcement learning and engineering capabilities [3] - The M2.5 model is fully integrated into the MiniMax Agent platform, refining core Office Skills capabilities and enabling users to create reusable industry experts, with over 10,000 custom experts now available covering high-frequency scenarios in office, finance, and programming [3]

Core Insights - The AI industry is experiencing significant upgrades in major models both domestically and internationally around the Spring Festival period in 2026, indicating a transformative phase in the sector [2][3]. Domestic AI Model Updates - Several domestic AI models have been updated recently: - January 27: Moonlight released the K2.5 model - February 12: Zhipu released the GLM-5 model - February 13: MiniMax released the MiniMax M2.5 model - February 14: ByteDance released the Doubao 2.0 model - February 16: Alibaba released the Qianwen 3.5 model [3]. International AI Model Updates - Internationally, major AI companies have also completed significant model updates: - OpenAI released the GPT-5.3-Codex model designed for real-time programming, capable of generating over 1000 tokens per second - Anthropic launched the Claude Opus4.6/Claude Sonnet4.6 models, supporting 1 million tokens context for better execution of intelligent planning tasks - Google introduced the Gemini3.1Pro model, which has doubled its inference performance [3]. Advancements in AI Applications - The capabilities of AI models are evolving, with the emergence of intelligent agents and multimodal applications: - The OpenClaw open-source intelligent agent tool has gained attention for its ability to assist with specific office tasks, moving beyond simple chat functionalities - ByteDance's SeeDance2.0 has improved video generation efficiency, increasing the usable rate of 15-second videos from 20% to 90%, potentially driving the anime and drama industries towards large-scale development [4]. Pricing Trends in Computing Power - A price increase in computing power has been observed, indicating a supply bottleneck in the computing power industry: - Zhipu's GLM Coding Plan has seen a price adjustment, including a 30% or more increase in package prices, reflecting rising market demand and user engagement [5]. Investment Recommendations - The computing supply chain is expected to experience sustained growth, with investment opportunities identified in: - Computing and storage: Cambrian, Haiguang Information, Moore Threads, Muxi, Zhaoyi Innovation, Baiwei Storage, Jucheng - Advanced manufacturing: SMIC, Huahong, Jinghe Integration, China Resources Micro, Chipone, Changdian Technology, Tongfu Microelectronics - Optical communication: Zhongji Xuchuang, Tianfu Communication, Xinyi Sheng, Changfei Optical Fiber, Zhongtian Technology, Hengtong Optic-Electric, Huaxin Electronics, Jiepte, Dazhu Laser [6][7].

Core Insights - The report by Frost & Sullivan indicates a significant increase in the daily usage of enterprise-level large models in China, with a rise to 37 trillion tokens in the second half of 2025, marking a 263% increase from 10.2 trillion tokens in the first half [1][3]. Group 1: Market Dynamics - The primary drivers for enterprises using large models shifted from enhancing product performance and customer experience in the first half of 2025 to improving operational efficiency and R&D effectiveness in the second half [3]. - The market for enterprise-level large models is rapidly consolidating around leading vendors, with the top three models (Qwen, Doubao, and DeepSeek) accounting for over 70% of daily token consumption in the second half of 2025, compared to less than half in the first half [3][4]. Group 2: Vendor Performance - Alibaba Cloud's Qwen model saw its market share nearly double to 32.1% in the second half of 2025, with a growth rate of 14.4% [3][4]. - The report highlights that enterprises are increasingly favoring fewer vendors to reduce long-term operational costs related to interface maintenance and version management, creating higher migration barriers for leading vendors [4]. Group 3: Open Source Trends - There is a notable increase in enterprises' willingness to adopt open-source models, with the proportion of closed-source users planning to add open-source options doubling from 22.6% in the first half to 48.5% in the second half of 2025 [4]. - Chinese companies accounted for 90.2% of the global new open-source large models introduced in the second half of 2025, significantly outpacing international competitors [4]. Group 4: Competitive Landscape - The competition among domestic open-source models has expanded beyond single-point performance to include size, deployment flexibility, and cost control [5]. - Since its launch in 2023, Qwen has released over 400 models, achieving over 1 billion downloads and extending its language capabilities to 201 languages, establishing itself as the leading open-source model globally [5].

"如今，我们已经实现了1.6T光模块的量产。"石川把光模块底部朝上，指着里面只有针尖大小的"点"介 绍，"这是光模块的通道，里面的8个'点'，每秒可传输1.6万亿位的数据。" 正月初七，湖北武汉光谷已开始忙碌起来。走进华工正源光子技术有限公司的无尘车间，机器轰鸣，灯 光通明。 春节不停歇，华工正源的信心来自市场的认可。"我们的订单已经排到2026年第四季度，现在工厂24小 时生产，每天两班倒。"华工正源AI（人工智能）数据通信项目生产经理石川拿起一个U盘大小的产品 告诉记者，"每天有3万支以上的光模块从这里发往全球。" "光模块负责电信号与光信号的转换，就像是数据跑动的'高速公路'。"石川介绍，光模块传输速率高， 数据处理就会快，AI的使用效果也会更好。 如此多的数据，要跑得又快又稳，这份底气来自哪里？石川拆出电路板，指着上面只有指甲盖一半大小 的器件说，"这些是光芯片，是光模块的'核心心脏'，它直接决定了光模块的传输速率。"石川说。 光芯片要发挥效率，要求并不简单。石川介绍，工厂要求每一块光芯片放置的位置，其上下左右的缝隙 不超过5微米。来到组装产线，只见工人们将组装有芯片的电路板放在显微镜上，一旁的电脑屏 ...