华夏时报记者 石飞月 北京报道 智谱新模型流量爆发，月之暗面模型调用量登顶全球，字节跳动视频生成工具震动好莱坞——春节前 夕，中国大模型厂商以罕见的密集节奏交出最新答卷。这一切有迹可循，DeepSeek去年春节前夕意外 破圈，所以2026年的同一个时间窗口也成为了国产大模型厂商心照不宣的"抢跑点"。 从这波新模型的特点来看，大模型的技术焦点已经从参数竞赛转向工程效率，从通用智能下沉至垂类落 地，而全球榜单上也出现了越来越多中国开源模型的名字。这意味着，市场的聚光灯不再只属于单一黑 马——这场没有硝烟的春节"卡位战"正悄然提速，为2026年大模型竞赛按下加速键。 春节前纷纷"交卷" 2月12日，上市仅一月出头的智谱股价大涨28.68%至402港元/股；2月13日，智谱涨势延续，股价截至收 盘涨20.65%至485港元/股，市值达到2162亿港元。 这是因为2月12日早间，智谱宣布上线并开源GLM-5，并称这款模型为"Agentic Engineering时代最好的 开源模型"，在真实编程场景的使用体感逼近Claude Opus 4.5，擅长复杂系统工程与长程Agent任务。 当然，仅仅发布一款新模型，资本市场不 ...

Investment Rating - The report maintains a "Hold" rating for the computer industry [6] Core Insights - The demand for Tokens is experiencing "inflation," which benefits cloud computing and gives model vendors pricing power [3][4] - The traditional internet model of free services is being disrupted as the industry shifts from free traffic to Tokens as a measurable unit of production [5][10] - The increase in Token consumption is driven by the evolution of user needs from simple Q&A to complex tasks requiring significant computational resources [8][10] Summary by Sections Token Demand Inflation - Token inflation refers to the structural increase in Token consumption per user over time, driven by more complex user interactions with models [8] - Users are increasingly utilizing models for tasks such as code reconstruction and document generation, leading to higher Token consumption [8][10] Changes in the Large Model Era - Tokens are becoming a measurable production resource rather than free traffic, with each interaction consuming computational resources [5][10] - The pricing strategy of model vendors is evolving, allowing them to convert computational scarcity into profit through tiered pricing and subscription models [10][11] Investment Recommendations - The report suggests monitoring the impact of price increases and Token demand on profit margins in the short term, while tracking subscription retention and expansion in the medium term [10][11] - Long-term prospects are positive for companies that can integrate AI into workflows, creating a demand for AI governance tools [11]

2月12日，上市仅一月出头的智谱股价大涨28.68%至402港元/股；2月13日，智谱涨势延续，股价截至收 盘涨20.65%至485港元/股，市值达到2162亿港元。 本报（chinatimes.net.cn）记者石飞月 北京报道 智谱新模型流量爆发，月之暗面模型调用量登顶全球，字节跳动视频生成工具震动好莱坞——春节前 夕，中国大模型厂商以罕见的密集节奏交出最新答卷。这一切有迹可循，DeepSeek去年春节前夕意外 破圈，所以2026年的同一个时间窗口也成为了国产大模型厂商心照不宣的"抢跑点"。 从这波新模型的特点来看，大模型的技术焦点已经从参数竞赛转向工程效率，从通用智能下沉至垂类落 地，而全球榜单上也出现了越来越多中国开源模型的名字。这意味着，市场的聚光灯不再只属于单一黑 马——这场没有硝烟的春节"卡位战"正悄然提速，为2026年大模型竞赛按下加速键。 春节前纷纷"交卷" "其实春节本来就是流量高峰与口碑传播的黄金窗口，今年发布尤其密集，其实是去年DeepSeek在春节 前破圈，厂商也想复制这个成功路径，为全年商业落地抢占先机。"张孝荣说。 这是因为2月12日早间，智谱宣布上线并开源GLM-5，并称这款模 ...

正如大家都看到，自1月9日登陆港股以来，MiniMax股价走出了令人惊叹的曲线。随着最新模型M2.5正式发布，今日（2月13日）MiniMax股 价再创新高，一度涨超13%，市值超过2000亿港元。 加上智谱，中国大模型仅仅开年就诞生了两家两千亿公司。 MiniMax市值冲破2000亿港元了。 如此，身后投资人迎来丰厚回报。其中最让人意外的要数米哈游了，这家游戏公司凭借着天使投资人的身份投进MiniMax，不仅精准卡位大模 型爆发的前夜，更是收获不菲回报——这笔投资账面赚了约100亿。 由此望去，这样的科技创富故事正在频繁上演。 MiniMax暴涨，天使轮回报100倍 从成立到上市仅用4年时间，核心管理层年纪不超过35岁。在中国大模型江湖，MiniMax是别具一格的存在。 春节将至，大模型圈已然杀疯。今日（2月13日）早上，MiniMax正式赴宴——官宣了其已经提前两天开跑的新模型M2.5，编程与Agent性能直 接对标Claude Opus 4.6级别。 网友们测试下来，纷纷直呼"极致轻量、极速推理，极低成本"。相当于，曾经要用大卡车拉的货物，如今用小板车就能运载完成。 市场迅速响应，MiniMax盘中市值 ...

视频生成模型引爆算力需求，字节跳动与智谱领跑 本周中国AI领域的进展令人目不暇接，各大厂商都在争夺技术高地。美银证券研报强调，字节跳动发布的Seedance 2.0不仅仅是一次迭代，它展示 了先进的多模态能力，特别是下一代视频生成技术。 与此同时，智谱AI推出了旗舰模型GLM-5，在编程和推理性能上实现了显著增强。DeepSeek更新了模型以支持更长的上下文窗口和更新的知识 库，MiniMax也在其海外平台上发布了M2.5版本进行测试。 这些模型的发布不仅仅是新闻通稿，它们直接催生了巨大的算力缺口。研报援引Fal.ai联合创始人的数据指出，生成一个5秒钟的视频片段（24帧/ 秒）所需的算力，大约是生成200个token（约150个字符）文本所需的12万倍。这一数据极具震撼力，意味着如果字节跳动的Seedance 2.0能够成功 推动视频生成的广泛应用，整个市场对Token消耗和算力基础设施的需求将出现数量级的飞跃。这对于拥有大量储备资源的IDC厂商来说，是基本 面的重大利好。 美银指出，中国AI行业本周迎来了极其关键的转折点。这不再仅仅是关于技术参数的军备竞赛，而是实打实的商业化落地与需求爆发。随着字节 跳动 ...

Investment Rating - The report maintains an "Outperform" rating for the industry, indicating a positive outlook compared to the broader market [3]. Core Insights - The AI application landscape is rapidly evolving, with significant advancements expected in 2026, which will impact traditional sectors such as software, film, and media [6]. - Domestic AI models are increasingly competitive with international counterparts, particularly in terms of performance and cost, suggesting a potential shift in the global AI model competition [6]. - The demand for AI inference is rising sharply, leading to increased pricing actions from model and cloud service providers, which benefits companies within the industry [6]. Industry Data Summary 1. Industry Data - The software industry in 2025 is projected to generate revenues of 15.48 trillion yuan, reflecting a year-on-year growth of 13.2% [12]. - The IC design sector is expected to grow by 18.9% in 2025, while cloud computing and big data services are anticipated to grow by 13.6% [19]. - The information technology service revenue is expected to increase by 14.7%, further solidifying its share of the overall software business revenue to 68.7% [27]. 2. AI Developments - Major AI models are being released, with significant competition expected in the market, particularly with the upcoming launches of models like DeepSeek-V4 and others in February 2026 [33]. - The pricing of domestic models is significantly lower than that of international models, enhancing their attractiveness for various applications [39]. - The user engagement for AI applications is increasing, with notable growth in active users for platforms like Gemini and ChatGPT [44]. 3. Domestic Market Dynamics - Domestic chip manufacturers are expected to benefit from restrictions on foreign competitors, with companies like Huawei and Dawning making significant advancements in AI computing capabilities [6]. - The report highlights the increasing capital expenditures from overseas tech firms, raising concerns about market dynamics [6]. - The MaaS (Model as a Service) market is projected to grow rapidly, with significant contributions from companies like ByteDance and Alibaba [56].

VC圈新贵。 作者/ 冯雨晨 杨继云 报道/投资界PEdaily Mi n iMa x市值冲破2 0 0 0亿港元了。 而米哈游，正是这一片盛景里最绕不开的存在。 MiniMax暴涨 天使轮回报100倍 从成立到上市仅用4年时间，核心管理层年纪不超过3 5岁。在 中国大模型江湖，Mi n iMa x是别具一格的存在。 春节将至，大模型圈已然杀疯。今日（2月1 3日）早上，Mi n iMa x正式赴宴——官宣了其已经提前两天开跑的新模 型M2 . 5，编程与Ag e n t性能直接对标Cl a u d e Op u s 4 . 6级别。 网友们测试下来，纷纷直呼"极致轻量、极速推理，极低成本"。相当于，曾经要用大卡车拉的货物，如今用小板车 就能运载完成。 市 场 迅 速 响 应 ， Mi n iMa x 盘 中 市 值 超 2 0 0 0 亿 港 元 。 如 此 君 子 豹 变 时 刻 ， Mi n iMa x 背 后 的 投 资 人 回 报 尤 为 引 人 关 注。 正如大家都看到，自1月9日登陆港股以来，Mi n iMa x股价走出了令人惊叹的曲线。随着最新模型M2 . 5正式发布， 今日（2月1 3日）M ...

生成式AI 一、智谱开源GLM-5，Agentic Engineering时代最强开源模型 1. 智谱发布开源GLM-5，参数规模扩展至744B（激活40B），在Artificial Analysis榜单位居全球第四、开源第 一，Coding与Agent能力逼近Claude Opus 4.5； 2. 模型在SWE-bench-Verified和Terminal Bench 2.0分别获得77.8和56.2分刷新开源SOTA，擅长复杂系统工程 与长程Agent任务； 3. GLM-5已完成与华为昇腾、寒武纪、昆仑芯等国产芯片适配，推出Z Code全流程编程工具和AutoGLM通用 Agent助手。 https://mp.weixin.qq.com/s/MVo6DIcGNje_YtLgVa4PaQ 二、MiniMax发布M2.5，10B激活参数达旗舰水平推理速度 更 快 1. MiniMax发布M2.5模型，激活参数仅10B却达到第一梯队旗舰水平，编程和Agent能力比肩Opus 4.6，推理速度 是Opus的3倍； 2. 实测9分钟搭完全栈学习网站，能独立完成物理模拟和企业级CMS系统搭建，支持PC/App/R ...

2025 年 DeepSeek 的 春 节 突 袭 战 术 已 成 为 行 业 教 科 书 。 2026 年 所 有 人 都 学 会 了 这 一 招 。 字 节 、 阿 里 、 智 谱 及 DeepSeek"春节档"密集发布旗舰模型，竞争焦点从单纯的模型性能转向"效率"与"智能体"落地。分析指，若DeepSeek V4实现推理降本，将推动AI从独立App向微信等高频应用"内嵌"转移，腾讯或成最大受益者。 硬·AI 作者 | 龙 玥 编辑 | 硬 AI 2026年的春节，不再仅仅是消费的狂欢，更演变成了中国AI巨头们争夺"移动端入口"的速度与激情。 据追风交易台消息，摩根大通在2月11日发布的最新研报中指出，中国互联网与AI行业正迎来史上最密集 的旗舰模型发布潮。这不再是单一模型的独角戏，而是一场关于谁能最快将"技术溢出"转化为"消费级爆 款"的抢凳子游戏。 01 拥挤的"春节发布季" 2025年DeepSeek的春节突袭战术已成为行业教科书。2026年，所有人都学会了这一招。 根据摩根大通的观察，市场再次为类似的"节假日发布季"做好了准备，但区别在于： "这次不是一枝独 秀，我们看到众多待发布的旗舰式和近旗舰 ...

Core Insights - The 2026 Spring Festival has transformed into a battleground for Chinese AI giants competing for "mobile entry points," with a focus on efficiency and the implementation of intelligent agents [1] - Morgan Stanley's report highlights an unprecedented wave of flagship model releases in the Chinese internet and AI sector, marking a shift from individual model performance to the rapid conversion of "technical spillover" into consumer-grade products [1][9] - DeepSeek remains the focal point of market attention, with potential cost reductions in inference that could shift AI from standalone apps to embedded applications in high-frequency platforms like WeChat [1][12] Group 1: Competitive Landscape - The Spring Festival release season is crowded, with ByteDance leading by unveiling three models: Seedance 2.0, Seedream 5.0, and Doubao 2.0, with Seedance 2.0 showing signs of becoming a "hit" [2] - Alibaba is set to launch Qwen 3.5 in mid-February, supported by a 3 billion yuan incentive plan to attract users [3] - Zhiyu released GLM-5 on February 11, expanding its parameter scale from 355 billion to 744 billion [4] - DeepSeek's V4 version is expected to be released in mid-February, focusing on improvements in encoding and handling long prompts [6] Group 2: Market Dynamics - The simultaneous release of multiple models may lead to a "winner-takes-all" scenario, where underperforming models face significant disadvantages [9] - The Spring Festival window is characterized by scarce attention, making it crucial for labs to present credible flagship updates to remain relevant [11] - DeepSeek's potential release could yield significant "platform economic benefits," with its new technology aimed at improving efficiency without heavy computational upgrades [11] Group 3: Implications for Major Players - Tencent is projected to be the biggest beneficiary of DeepSeek's potential new releases, as it integrates third-party model capabilities into its high-frequency communication platforms [13][14] - Stronger models could enhance user experience for Alibaba and Baidu, but a price war initiated by DeepSeek could pressure the entire industry's API services [15] - Vertical giants like Ctrip, Beike, and Kuaishou stand to benefit from powerful open-source models that lower technical barriers and accelerate product iteration [15] Group 4: Future Outlook - The upcoming large-scale user testing during the Spring Festival will serve as a critical test for the reliability and performance of consumer-grade AI applications [17] - Morgan Stanley maintains a bullish outlook on model developers Zhiyu and MiniMax, with target prices set at 400 HKD and 700 HKD respectively, based on projected earnings by 2030 [18][19] - The valuation logic is shifting towards economic benefits, focusing on stronger willingness to pay and higher API workload retention rates as model capabilities approach global frontiers [19]