除夕夜央视春晚上被网友点赞"每一帧都好美"的节目《贺花神》，如同"花"的视觉盛宴。殊不知，这属 于中国人的美学浪漫背后有大模型的助力。2月12日，北京企业字节跳动推出视频生成模型Seedance 2.0。"今年的总台春晚是Seedance 2.0的第一个客户。"火山引擎总裁谭待说，正式上线前，该模型就被 深度应用于多个春晚节目的视觉制作环节。 当中国观众享受春晚的视觉特效时，像Seedance 2.0这样诞生在北京的国产大模型正引发海外开发者和 创作者的疯狂追捧。今年以来，全球迎来新一代大模型的密集发布潮。北京在中国AI企业中一马当 先、万"码"奔腾。 大模型上春晚 解决海外模型解不了的难题 《贺花神》节目使用Seedance 2.0模型，为每位演员的表演定制了"一月一人一景，一花一态一观"的视 觉奇观，让舞台变成可呼吸、可生长的动态空间。歌手张杰演唱《驭风歌》时，身后水墨画上的骏马活 灵活现。不过，水墨风格的素材如此灵动，考验模型对东方美学的理解和泛化能力。此前，一些海外模 型由于不理解中国水墨画构图和逻辑，最终生成的视频出现了变形。 "训练阶段我们用了大量东方美学素材，而且它在镜头语言的控制、复杂交互呈现 ...

文｜周鑫雨 编辑｜苏建勋 春节，热钱涌动在中国大模型赛道上。 一端，是狂撒几十亿元发红包、请奶茶，替当家AI应用拉流量的大厂们；另一端，则是股价飞涨、融 资迅猛的大模型六小虎们。 在搞钱的战场上，开年最大的一笔融资，来自月之暗面。 此前彭博社等媒体报道称，近期月之暗面即将完成的超7亿美元融资，由阿里、腾讯、五源资本、九安 医疗等老股东领投，并且已经超募。与此同时，月之暗面已经以100亿-120亿美金的估值，无缝开启了 新一轮融资。 《智能涌现》独家获悉，月之暗面的超7亿美元融资中，股东阵容除了联合领投的老股东们，还包括老 股东高榕创投，以及新增的凯辉基金——这也是凯辉，首次朝大模型公司开枪； 据我们了解，以超100亿美金估值开启的新一轮融资，已经收到了多家机构的意向，包括欧洲背景的海 外基金。 有关上述信息，截至发稿前，月之暗面暂无回应。 用"抢"形容月之暗面这轮融资的局面，一点也不过分。 一名知情人士告诉我们，从2025年下半年，月之暗面开启融资之初，不少机构的LP，就"催着投资"。 这或许是除了2023年初，行业对大模型最大的一次FOMO（错失恐慌症）。 溯其缘由，是自港股IPO以来，智谱和MiniMax ...

IT之家 2 月 18 日消息，今日特斯拉、X 平台 CEO 埃隆 · 马斯克通过 X 平台发文官宣，xAI 旗下大模型 Grok 的 4.2 版本候选发布版（公开测试版）现已可供 使用。 IT之家注意到，这位全球首富在 X 平台上写道，用户必须专门选择 Grok 4.2 才能激活它。 马斯克在推文中表示，非常欢迎用户为该版本提供反馈。他同时强调，与 Grok 此前的所有版本不同，Grok 4.2 具备快速学习能力，因此每周都会有带有发 布说明的改进。 ...

Core Viewpoint - The company is expanding its international financing channels through the issuance of H-shares and plans for a listing on the Hong Kong Stock Exchange by December 15, 2025 [1] Financial Performance - In the first half of 2025, the company achieved a revenue of 556 million yuan, representing a year-on-year growth of 20.40%, primarily driven by its content ecosystem and IP operations [2] Business Progress - As of August 22, 2025, the domestic box office for a specific film has surpassed 460 million yuan, receiving high ratings. The company plans to release this film in multiple countries and regions, including North America, Southeast Asia, and Japan starting in August 2025 [3] Business and Technology Development - The overseas short video app FlareFlow, launched in April 2025, has rapidly expanded to cover 177 countries and regions, with recharge revenue increasing by over 500% within three months. Additionally, the company's self-developed "Chinese Xiaoyao" large model has been applied in content creation to enhance production efficiency [4]

Core Insights - Kimi, a product of Moonshot AI, has successfully raised over $700 million in a new financing round, following a $500 million Series C round just a month prior, bringing its valuation to over $10 billion and marking its entry into the global unicorn club [1][6] - The total financing for Kimi in the last two months exceeds $1.2 billion, setting a record for the domestic large model industry in nearly a year [7] - The Chinese AI industry is transitioning from a phase of numerous models to a more consolidated market, with the number of capable and well-funded model developers decreasing from over 200 to less than 10 [9][10] Company Developments - Kimi's CEO, Yang Zhilin, indicated that the company has over 10 billion RMB in cash reserves and is not in a hurry to go public, as it can raise more funds from the primary market compared to the secondary market [7][10] - Kimi's global paid user base is growing at a monthly rate of 170%, and its API revenue from overseas has increased fourfold, driven by its K2 Thinking model [10] Market Trends - The competition in the large model sector is shifting from a focus on technology to the ability to build commercial systems [9][10] - The recent IPO of Zhiyu, the first company focused on general artificial intelligence (AGI) in China, raised over 4.3 billion HKD with a subscription rate of 1159.46 times for public offerings, indicating strong market interest [8] - MiniMax, another AI large model company, has also recently listed on the Hong Kong Stock Exchange, with its stock price rising by 65% in the first three trading days [8]

Group 1 - The core viewpoint of the article is that there is a coordinated effort by marketing accounts to portray ByteDance as the largest short-seller of Hengke, misinterpreting coincidences as causal relationships, which is seen as malicious and absurd [2] - Li Liang, Vice President of Douyin Group, acknowledges that while some decline in Hengke's stock may be related to Seedance 2.0, certain key stocks in Hengke's computing power sector are experiencing gains due to the emergence of better models in China [2] - The impact of Seedance 2.0 on Hengke is multifaceted and should be viewed comprehensively, as innovations and competition in large models will ultimately drive the development of AI technology, benefiting all participating companies through mutual stimulation [2]

Core Insights - Alibaba has launched its new AI model Qwen3.5-Plus, which competes with top closed-source models like Gemini-3-pro and GPT-5.2, marking a significant advancement in the AI landscape [1] - During the Spring Festival, over 130 million people experienced AI shopping for the first time, with the AI assistant "Qianwen" becoming a national-level tool [1] - Qwen3.5-Plus features a total of 397 billion parameters, with only 17 billion activated, achieving performance that surpasses models with over a trillion parameters while reducing deployment memory usage by 60% [1][4] Model Performance and Innovation - Qwen3.5 represents a generational leap from previous models, transitioning from a pure text model to a native multimodal model, significantly enhancing its capabilities in various assessments [2] - The model has shown superior performance in multiple evaluations, including general agent assessments and visual question answering, outperforming competitors like Gemini 3 Pro and GPT-5.2 [2][3] - Key innovations in Qwen3.5's architecture include a mixed MoE structure and a unified approach to handling multiple data modalities during pre-training, resulting in a more efficient model with lower parameter requirements [4] Market Impact and User Engagement - The Qianwen app has seen a dramatic increase in user engagement, with a 22-fold increase in AI ticket orders and a 372-fold increase in movie ticket orders during the Spring Festival [6] - Nearly half of all AI orders came from county-level cities, indicating a broadening user base and accessibility of AI technology [6] - The app's integration with Qwen3.5-Plus enhances its capabilities, allowing it to assist users in various tasks, thus increasing its utility in daily life [6] Infrastructure and Ecosystem - Alibaba has opened over 400 Qianwen models since 2023, with global downloads surpassing 1 billion, fostering a robust developer ecosystem [8] - The company’s self-developed AI chip "Zhenwu 810E" is optimized for training and inference of large models, contributing to the cost-effectiveness of Qwen3.5 [9] - Alibaba Cloud holds a 35.8% market share in China's AI cloud market, indicating its leading position and the growing demand for AI infrastructure [10] Long-term Strategy and Vision - Qwen3.5 is part of Alibaba's broader strategy towards achieving AGI, with ongoing exploration of all-in-one multimodal capabilities [5][7] - The dual approach of advancing cutting-edge models while making AI accessible to everyday users reflects Alibaba's commitment to integrating technology into daily life [11]

Group 1 - Kimi is set to complete a new financing round of over $700 million, following a previous $500 million round just over a month ago [1] - The latest financing round is led by existing shareholders including Alibaba, Tencent, Wuyuan, and Jiua [1] - Kimi's valuation has doubled to exceed $10 billion, with the total amount raised in the last two rounds reaching over $1.2 billion, marking the highest financing amount in the large model industry in the past year [1]

此外，其新一轮 100-120 亿美元（现汇率约合 691.52 - 829.82 亿元人民币）估值的融资亦已经开启，照此计算，Kimi 最新估值直接翻倍并突破 100 亿美元 大关，总额超 12 亿美元（IT之家注：现汇率约合 82.98 亿元人民币）的连续两轮融资也是近一年来大模型行业的最高融资金额。 IT之家 2 月 17 日消息，据"科创板日报"报道，在完成上一轮 5 亿美元（现汇率约合 34.58 亿元人民币）融资仅一个多月后，月之暗面 Kimi 新一轮超 7 亿美 元（现汇率约合 48.41 亿元人民币）的融资即将完成交割，本轮由阿里、五源、九安等老股东联合领投，腾讯也参与投资。 ...

Core Insights - Kimi, a subsidiary of Moonshot AI, is set to complete a new financing round exceeding $700 million, following a previous $500 million round just over a month ago, with a valuation expected to reach between $10 billion and $12 billion [1][6] - The company has achieved a significant milestone with its latest valuation surpassing $10 billion, marking the highest financing amount in the large model industry in the past year [1][6] - Kimi's founder, Yang Zhilin, indicated that the company is not in a hurry to go public, preferring to raise more funds from the primary market rather than the secondary market [3][9] Financing and Valuation - The recent financing round is led by existing investors including Alibaba, Wuyuan, and Jiua, with Tencent also participating [1][6] - Kimi's total cash reserves have exceeded 10 billion yuan, providing a strong financial foundation for future growth [1][6] Technological Advancements - Kimi has made significant technological strides, achieving "SOTA" (state-of-the-art) results, with the release of K2 and K2 Thinking marking important steps towards AGI [3][9] - The company has launched multiple agent functionalities since May, including Researcher, OK Computer, PPT, and Kimi Code, contributing to a 170% month-over-month growth in paid users from September to November [3][9] Strategic Focus - The funds from the recent financing will be used to aggressively expand GPU resources and accelerate the training and development of the K3 model [4][10] - Kimi plans to enhance employee incentives significantly in 2026, aiming to double the average incentives compared to 2025 [4][10] Market Trends - The demand for AI talent has surged, with a tenfold increase in job openings for AI positions in the first seven months of 2025, highlighting a competitive landscape for talent acquisition [5][10] - Kimi's strategic goals for 2026 include improving the K3 model's performance, integrating model training with product development, and focusing on maximizing productivity rather than just user numbers [5][10][11]