7月8日笔者独家获悉，年初爆火的AI Agent公司Manus（北京蝴蝶效应科技有限公司）近期对国内进行 大规模裁员，120名员工中，除40余名核心技术人员迁往新加坡总部，其余皆被裁员，并且将总部从中 国迁至新加坡。 作为全网首个发布Manus裁员消息的人，笔者万万没有想到，这一消息最终会引发如此广泛的讨论。 过去五天里，数亿人次在多个社交平台及 AI 行业圈进行热议。彭博社（Bloomberg）、The information 等多家外媒也对此事进行报道。 事实上，凭借"全球首个通用AI智能体"产品惊艳亮相仅不到四个月，如今，Manus公司不仅选择大规模 裁员，而且将微博、小红书等国内社交平台清空，官网屏蔽一切中国IP访问"Manus在你所在的地区不 可用"，并全面删除与阿里的合作承诺。 这家AI Agent新贵最终以最极端的方式"宣告"其下一步的计划：毅然离开中国，Manus将成为一家新加 坡公司。 复盘整个事件，Manus最终"塌房"并陷入国内舆论旋涡的点在于，"昙花一现"的Manus并不像 DeepSeek一样覆盖国外、国内市场，而是在中美AI博弈之下，以及字节、百度等大厂都在发力AI Agent之际 ...

截至发稿，Manus方面尚未作出回应。 此前回应裁员消息 7月11日，记者发现，登录Manus官网，已显示为"Manus在你所在的地区不可用"，此前为"Manus中文版本正在开发中"。 同时，Manus的官方微博和小红书账号已清空内容，只有微信视频号ManusAI留下一则宣布"与阿里通义千问团队正式达成战略合作"的信息。 有消息称Manus对旗下部分国内业务进行裁员，并将核心技术人员迁往新加坡总部。 不过，Manus的海外社交平台X上的账号依旧保持更新，该账号显示位置为"新加坡"，最新一条更新时间为7月10日，介绍一场定于13日在旧金 山的徒步活动信息。 硅谷风投领投B轮融资 Manus创始人团队包括肖弘、季逸超和张涛三位。 肖弘毕业于华中科技大学，先后做过微伴助手和壹伴，2022年时曾创立北京蝴蝶效应科技有限公司，推出AI浏览器插件Monica。季逸超毕业于 北京信息科技大学，2012年曾创立Peak Labs，推出Rasgueado输入法，获红杉资本和真格基金投资。张涛曾任神策数据副总裁、字节跳动国际 化负责人等。 今年3月"一炮走红"后，Manus在业务与融资方面的消息频传。 3月11日，Manus与阿 ...

作者 | 梁昌均 编辑 | 杨锦 近日，此前全球爆火的AI Agent（智能体）Manus，因国内裁员、迁址新加坡引发行业关注。 媒体报道称，Manus国内进行裁员，其在中国区的员工总数120人左右，除了40多名核心技术人员迁往新加坡总部之后，其余员工都将会进行裁员优化，给 予N+3或者2N的赔偿。 搜狐科技7月11日实地探访了Manus北京办公区海淀区epark花园路社区1号共享大厅，该园区入驻的大多都是科技类公司。据多位在此办公的企业人士介绍， Manus办公区在四层，公司门口没有Logo，目前大概还有十几个人在此办公。 出品 | 搜狐科技 此外，还有媒体援引Manus离职员工称，Manus与阿里通义千问的中文版合作开发事宜将不会继续推进。 今年3月11日，Manus宣布，双方将基于通义千问系列开源模型，在国产模型和算力平台上实现Manus的全部功能。"两家技术团队已展开紧密协作，致力于 为中国用户打造更具创造力的通用智能体产品。" 对于停止合作消息，阿里云内部人士向搜狐科技表示，这个消息是个乌龙，该员工纯粹是个人猜测，Manus官方没说过这话。他还强调，"千问是开源模 型，就算他们真搬去新加坡了，未来也 ...

Core Viewpoint - Manus, a general AI agent company, has cleared its social media accounts and its website indicates unavailability in certain regions, raising concerns about its operations in China [1][4]. Group 1: Company Operations - Manus has reportedly laid off a portion of its workforce in China, with around 120 employees remaining, and has moved core technical staff to its headquarters in Singapore [4][5]. - The company stated that the layoffs were part of a strategy to improve operational efficiency and focus on core business development [4]. Group 2: Strategic Partnerships - In March, Manus announced a strategic partnership with Alibaba's Tongyi Qianwen team to develop a Chinese version of its product based on open-source models [4][5]. - Recent reports suggest that the collaboration on the Chinese version may not continue, although confirmation from both Manus and Alibaba has not been received [3][5]. Group 3: Product Launch and Market Response - Manus gained significant attention in March when it launched its product, which was marketed as the first true general AI agent capable of independent thinking and task execution [5]. - The demand for access to Manus was so high that invitation codes were being sold for as much as 100,000 yuan on secondary markets, despite the company stating it had not established any paid channels for obtaining invitations [5].

Manus方面对此回应智通财经称，"基于公司自身经营效率考量，我们决定对部分业务团队进行调整。 公司将继续专注核心业务发展，提升整体运营效率。" Manus由中国初创公司蝴蝶效应（Butterfly Effect）在今年3月6日发布，号称"全球首款通用智能体产 品"。当时该产品采用邀请内测的方式向公众开放，邀请码一度被炒至上万元，因此在收获盛赞的同时 对其炒作及套壳的质疑声也不断。 4月下旬旋即传出Manus母公司蝴蝶效应获硅谷知名风投Benchmark领投，融资7500万美元（约合5.5亿 元人民币）的消息。 21世纪经济报道记者 陈归辞 上海报道 今年3月爆火的AI智能体平台Manus传出裁员消息，仅4个月不到就将总部迁到新加坡。 近日，有消息称Manus对旗下部分国内业务进行裁员，并将核心技术人员迁往新加坡总部。 3月11日，Manus与阿里通义千问团队宣布达成战略合作，计划基于通义千问系列开源模型，在国产模 型和算力平台上实现Manus的全部功能。3月28日，其公布收费方案，开启商业化进程。5月13日， Manus正式开放注册。6月4日，其推出文本转视频服务。 融资方面，4月下旬，Manus母公司据悉 ...

Manus在今年3月发布AI助手早期预览版时，曾被称为"全球首款通用智能体产品"。用户仅需提供简单指令，该产品即可自动完成复杂任务。当时由于需要 内测邀请码才能体验，在社交平台掀起"求邀请码"热潮。5月13日，Manus宣布开放注册，为新用户提供1000积分奖励，每天可免费获得300积分执行任务。 在收费策略方面，Manus推出三档付费计划：每月19美元的基础计划、39美元的Plus计划和199美元的Pro计划。然而，许多用户认为其定价偏高。今年6月， Manus产品负责人张涛公开表示，公司已将总部从中国迁至新加坡，标志着公司国际化战略的重要转变。 组织架构重新配置 Manus背后的公司为ButterflyEffect（蝴蝶效应），该公司于2023年在香港成立，对北京红色蝴蝶科技有限公司拥有100%实际控制权。后者由肖弘创立于 2022年，曾推出主攻海外市场的AI浏览器插件Monica。今年5月，有外媒报道称，蝴蝶效应在由美国风险投资公司Benchmark领投的融资中筹集了7500万美 元，估值达到5亿美元，但该投资目前正接受美国财政部审查。 近期，AI智能体平台Manus因裁员传闻引发业界关注。该公司正式回 ...

Core Viewpoint - The article presents a new caching acceleration solution called HarmoniCa, which addresses the slow inference speed and high costs associated with diffusion models, particularly the Diffusion Transformer (DiT) architecture, achieving high-performance lossless acceleration [1][7][30]. Group 1: HarmoniCa Framework - HarmoniCa is designed to overcome the speed bottlenecks of the DiT architecture during deployment, enabling efficient training and inference collaboration through a feature caching acceleration framework [1][30]. - The framework introduces two key mechanisms: Step-Wise Denoising Training (SDT) and Image Error Proxy Objective (IEPO), which align training and inference processes to enhance performance [10][15][16]. Group 2: Mechanisms of HarmoniCa - SDT simulates the entire denoising process during training, reducing error accumulation and improving final image clarity and stability [11][12][15]. - IEPO focuses on optimizing the final image quality rather than intermediate noise errors, ensuring that the training objectives align with the ultimate goal of high-quality image generation [15][16]. Group 3: Experimental Results - HarmoniCa was tested against various methods, including Learning-to-Cache (LTC) and heuristic caching methods, demonstrating superior performance in terms of image quality and acceleration [17][19][20]. - In high compression scenarios (10-step inference), HarmoniCa maintained image quality advantages, achieving lower FID scores compared to LTC while improving cache utilization [19][22]. Group 4: Comparison with Other Techniques - HarmoniCa outperformed traditional pruning and quantization methods, providing stable acceleration without relying on specialized hardware, making it a more versatile deployment option [21][24]. - The framework showed compatibility with quantization techniques, enhancing inference speed while maintaining image quality, indicating its potential as an "acceleration plugin" [24][25]. Group 5: Cost Analysis - HarmoniCa demonstrated significant advantages in both training and inference costs, with a training time reduction of approximately 25% compared to mainstream methods, and minimal impact on throughput during inference [27][28]. - The lightweight nature of the added Router in the inference process ensures that it occupies only 0.03% of parameters, contributing to its efficiency [28]. Group 6: Conclusion - The article concludes that the HarmoniCa framework represents a new paradigm in caching acceleration, emphasizing the importance of synchronized training and inference processes to achieve optimal performance, efficiency, and adaptability in real-world deployments [29][30].

行 华福证券 潮玩：LABUBU 二手价格基本企稳，关注暑期旺季带动板块营收预期 提升。 行业监管风险、二手市场价格波动等因素共振导致估值回调：《人 民日报》点名盲盒盲卡乱象引发市场担忧，泡泡玛特官方大量补货后 本周二手交易价格基本企稳，泡泡玛特周波动幅度 7.6%，布鲁可为 -3.9%。根据 Google Trend 近 90 日搜索指数，当前 LABUBU 热度已超 过任天堂 Switch2 和 LOL（英雄联盟），在流行文化中表现强势。长 期看好中国软实力提升带动文化消费品出海，海外扩张与新兴市场消 费升级仍存较大赛道红利。 黄金珠宝：建议关注终端店效高增及门店拓展空间大的品牌。 金价持续上涨，一口价产品增长表现较好，面向高端客群及年轻 客群的品牌终端店效有望持续高增且门店拓展空间大，建议关注潮宏 基、莱绅通灵、老铺黄金；传统黄金批发品牌此前多季度出货下滑， 消费者逐步接受高金价后克重黄金消费量有望企稳，终端库存消化后 预计 Q2 陆续补货，主要品牌出货有望改善，建议关注老凤祥、周大福、 周大生；北京地区金银珠宝社零数据高增，建议关注北京区域龙头且 纯直营模式的菜百股份。 华福证券 社会服务 2025 ...

Core Viewpoint - Ying Tong Holdings has successfully listed on the Hong Kong Stock Exchange, becoming China's "first fragrance stock" within a year of submitting its prospectus [1]. Financial Performance - Ying Tong Holdings reported revenues of 1.699 billion yuan, 1.864 billion yuan, and 2.083 billion yuan for the fiscal years ending March 31, 2023, 2024, and 2025, respectively. Net profits for the same periods were 173 million yuan, 206 million yuan, and 227 million yuan [4]. - The company has shown continuous revenue growth, primarily driven by its agency business model, which heavily relies on fragrance brand representation [4]. Business Model and Brand Dependency - The company has secured agency rights for over 60 international fragrance brands, including Hermes, Chopard, Albion, Dolce & Gabbana, and Lola Marc Jacobs. Fragrance sales accounted for 89.3%, 88.5%, and 81.7% of total revenue for the fiscal years 2022 to 2024 [4][5]. - However, the heavy reliance on agency business poses a risk, as many beauty companies are reclaiming brand agency rights. Notably, Kering Group and Richemont Group have announced plans to take back control of their fragrance brands [6]. Brand Authorization Risks - As of March 31, 2025, Ying Tong Holdings has 65 external brands in its portfolio, with 4 brands no longer under its authorization. Among the remaining brands, 22 will see their authorization expire within a year, and 19 within one to three years [6]. - The loss of agency rights has already impacted the company's gross margin, which fell from 48.3% in fiscal 2023 to 45.8% in fiscal 2024 due to a non-renewed distribution agreement with a luxury brand [7]. Market Competition and Funding Needs - The competitive landscape in the cosmetics industry is intensifying, with rising price competition and marketing expenditures. This necessitates additional funding for companies to enhance market competitiveness, making IPO financing a crucial avenue for Ying Tong Holdings [7]. Strategic Initiatives - To mitigate risks and enhance growth, Ying Tong Holdings plans to use part of the raised funds to develop its own brands and invest in external brands. The company has already launched its own eyewear brand, Santa Monica, and entered the fragrance market in 2022 with five products [8]. - Currently, the revenue from self-owned products accounts for less than 1% of total revenue, indicating significant room for growth in this area [8].

中国经济网北京6月26日讯 颖通控股有限公司（以下简称"颖通控股"，06883.HK）今日在港交所上市， 开盘即破发，报2.58港元。截至今日收盘，颖通控股报2.40港元，跌幅16.67%。 颖通控股拟将所得款项净额用于"进一步发展自有品牌（包括SantaMonica），以及收购或投资外部品 牌""发展及扩大直销渠道，包括扩展在线及线下拾氛气盒门店以及其他自营在线及线下门店�u专柜""加 速数字化转型，主要方式是升级数字化CRM系统、中台系统以及财务及运营系统""提升集团的知名度 及声誉""拨付营运资金及一般企业用途，以支持业务营运及增长"。 颖通控股本次发行的联席保荐人为法国巴黎证券（亚洲）有限公司、中信证券（香港）有限公司，保荐 人兼整体协调人、整体协调人、联席全球协调人、联席账簿管理人及联席牵头经办人为法国巴黎证券 （亚洲）有限公司、中信里昂证券有限公司，整体协调人、联席全球协调人、联席账簿管理人及联席牵 头经办人为招银国际融资有限公司、星展亚洲融资有限公司，联席账簿管理人及联席牵头经办人为中湾 国际证券有限公司、第一上海证券有限公司、富途证券国际（香港）有限公司、软库中华金融服务有限 公司。 截至20 ...