事关Manus被收购一事，商务部发声！ 1月8日，在商务部例行新闻发布会上，新闻发言人何亚东在回应有关审查Meta收购人工智能平台Manus 的提问时说，中国政府一贯支持企业依法依规开展互利共赢的跨国经营与国际技术合作。需要说明的 是，企业从事对外投资、技术出口、数据出境、跨境并购等活动，须符合中国法律法规，履行法定程 序。商务部将会同相关部门对此项收购与出口管制、技术进出口、对外投资等相关法律法规的一致性开 展评估调查。 交易金额或在40亿—50亿美元 在去年3月份，由武汉创业公司"蝴蝶效应"研发的智能体Manus曾引起科技圈轰动。 3月6日凌晨，"蝴蝶效应"研发的全球首款通用型AI Agent Manus正式上线内测，凭借"独立思考—规划 —执行"的全链路闭环能力引爆科技圈。官网因访问量暴增一度瘫痪，二手交易平台上的内测邀请码更 被炒至5万元天价，创下AI行业新纪录。 在2025年年末，市场传Manus被Meta收购的消息，据《晚点LatePost》报道，此次Meta收购Manus的费 用为数十亿美元，也有业内人士估计，此次交易金额或在40亿—50亿美元。 12月30日，硅谷科技巨头Meta突然宣布"Ma ...

谁能想到，AI界最权威的大模型排行榜，竟然是个彻头彻尾的骗局？ 最近，2025年底的一篇名为《LMArena is a cancer on AI》的文章被翻了出来。 登上了Hacker News的首页，引起轩然大波！ 炸裂的是，这篇文章直接把LMArena——这个被无数研究者奉为圭臬的评测平台——钉在了耻辱柱上，称其为AI发展的「癌症」。 从金标准到毒瘤 所以，LMArena究竟是何方神圣？ 先说说背景。 LMArena（也叫LMSYS Chatbot Arena）是由加州大学伯克利分校、卡内基梅隆大学等顶尖学府的研究者于2023年创建的大模型评测平台。 | 图 Text | | | 1 8 days ago | WebDev | | | 1 9 days ago | | --- | --- | --- | --- | --- | --- | --- | --- | | Rank 14 | Model 11 | Score J | Votes 1 | Rank 11 | Model 1J | Score ↓ 0 | Votes 11 | | 1 | G gemini-3-pro | 1490 | 21,938 ...

Meta收购Manus，或许还未到尘埃落定之际。 2026年1月8日，就在外界以为Meta收购Manus一事已尘埃落定之际。 中国商务部在例行新闻发布会上的一番表态，为这笔备受瞩目、金额达数十亿美元的交易增添了新的不确定性。 据商务部新闻发言人何亚东透露，商务部将会同相关部门对此项收购与出口管制、技术进出口、对外投资等相关法律法规的一致性开展评估调查。 消息链接：商务部回应审查Meta收购Manus 这意味着，这笔被视为AI领域里程碑式的收购，或许依然处于「悬而未决」的状态。 商务部回应：评估调查正在进行 商务部的回应措辞审慎。 何亚东表示，中国政府一贯支持企业依法依规开展互利共赢的跨国经营与国际技术合作，但同时强调，企业从事对外投资、技术出口、数据出境、跨境并 购等活动，须符合中国法律法规，履行法定程序。 值得注意的是，商务部使用的措辞是「评估调查」，而非「正式调查」。 通俗解释一下，就是中国有一份「技术出口清单」，把技术分成三类： 这一表述存在解读空间：既可能是常规的合规审查，也可能升级为更深入的调查程序。 商务部回应涉及的法律法规解读 商务部在回应Manus收购案时，提到了「出口管制、技术进出口、对外投 ...

美国科技巨头Meta，要斥资140亿元人民币，收购中国AI初创公司Manus的事，有可能发生变故。 2025年3月，Manus以"全球首款通用Agent"的身份，带来行业狂热，之后陷入"套壳""不可控"的质疑声浪，再到4月完成7500万美元融资、估值逼近5亿美 元，直到被Meta看中 。 1月8日，商务部新闻发言人何亚东，在回应有关审查Meta收购Manus的提问时说，中国政府一贯支持企业依法依规开展互利共赢的跨国经营与国际技术合 作。 值得一提的是，Manus一开始就瞄准了全球化。2025年7月，在爆火4个月后，Manus将总部搬到了新加坡，并裁掉了国内七成团队。 同时，企业从事对外投资、技术出口、数据出境、跨境并购等活动，须符合中国法律法规，履行法定程序。商务部将会同相关部门对此项收购与出口管制、 技术进出口、对外投资等相关法律法规的一致性开展评估调查。 目前看来，这一收购，有更复杂的情况，可能通向不同的结果。 去年12月30日，Meta宣布完成对Manus的收购。这场由CEO扎克伯格亲自操盘的谈判仅用十余天敲定，交易对价超过20亿美元（约合人民币140亿元），跻 身Meta史上第三大并购案。若收购完成， ...

Core Viewpoint - Beijing Zhiyu Huazhang Technology Co., Ltd. (Zhiyu) officially listed on the Hong Kong Stock Exchange on January 8, becoming the "first global large model stock" with a closing price increase of 8.06%, reaching a market capitalization of over HKD 55 billion [1] Group 1: Company Performance - Zhiyu's public offering was oversubscribed by 1,159.46 times, indicating strong market interest, with cornerstone investors committing a total of HKD 2.98 billion [1] - Despite a volatile first trading day, the stock price fluctuated but ultimately stabilized above the issue price [1] - The company reported a significant revenue increase from CNY 57.4 million in 2022 to CNY 312.4 million in 2024, but also faced a cumulative net loss of approximately CNY 6.2 billion from 2022 to mid-2025 [2] Group 2: Business Model and Strategy - Zhiyu, founded in 2019, has a strong technical foundation stemming from Tsinghua University's Computer Science Department, focusing on making machines think like humans [3] - The company primarily generates revenue by providing localized deployment of large model services, accounting for about 70% of total revenue [4] - Zhiyu plans to allocate approximately 70% of the net proceeds from its IPO (around HKD 2.9 billion) to further develop general AI large models, aiming to strengthen its technological moat [2] Group 3: Industry Outlook - The large model industry is experiencing rapid growth, with domestic companies in China leading in global competition, although no single company has established a dominant market position yet [4] - The industry faces challenges such as high research and development costs and increasing losses, which are common among AI large model companies [2]

智源研究院理事长黄铁军分享了他的技术趋势观察：AI的发展要重视"结构决定功能，功能塑造结构"的相互作用。当前人工智能正从功能模仿转向理解物理 世界规律，这一根本转变意味着AI正褪去早期狂热，其发展路径日益清晰，即真正融入实体世界，解决系统性挑战。 随后，智源研究院院长王仲远发布了十大AI技术趋势，详细阐释了这一变革。基础模型的竞争，焦点已从"参数有多大"转变为"能否理解世界如何运转"。他 指出：我们正从 "预测下一个词"跨越到"预测世界的下一个状态"。这标志着以"Next-State Prediction"（NSP）为代表的新范式，正推动AI从数字空间的"感 知"迈向物理世界的"认知"与"规划"。 报告认为，2026年将是AI从数字世界迈入物理世界、从技术演示走向规模价值的关键分水岭。这一转变由三条清晰的主线驱动： 首先，是认知范式的"升维"。以世界模型和NSP为核心，AI开始学习物理规律，这为自动驾驶仿真、机器人训练等复杂任务提供全新的"认知"基础，成为国 内外领先模型厂商竞相布局的战略高地。 其次，是智能形态的"实体化"与"社会化"。智能正从软件走向实体，从单体走向协同。头部科技公司的人形机器人正进入真实 ...

Group 1 - The Chinese commerce ministry is currently reviewing Meta Platforms' acquisition of Manus, which is valued at over $2 billion [1] - Manus is an artificial intelligence start-up based in Singapore and has Chinese founders [1]

来源：环球网 当地时间1月6日，2026年国际消费电子展（CES 2026）在美国拉斯维加斯正式举办。作为专业的大模 型对话式人工智能平台型企业，思必驰展示了全链路语音语言交互技术在智慧出行领域的最新突破，彰 显中国AI技术在全球化舞台上的竞争力。 全场景交互突破：天琴语音助手海外版 本次展会上，思必驰带来基于全链路智能语音语言交互技术的天琴语音助手海外版。该系统集成思必驰 自主研发的端到端语音技术，涵盖语音处理、识别、合成、语言理解、噪声抑制、口语容错、对话管 理、全双工架构等技术，可为全球车企客户快速实现对话系统的柔性定制，满足不同地区用户的座舱交 互需求。 此外，该系统实现了全双工免唤醒、一句话多意图理解、可见即可说等创新功能，显著提升了驾驶安全 性与交互自然度。声纹识别与多音区声源定位技术的加持，让不同座舱位置的乘客都能获得个性化响 应，构建起真正意义上的全场景智能交互体验。 在思必驰展位的体验区，参观者通过搭载多语种语音系统的智能座舱设备，亲身体验了接近真人对话的 交互效果。系统在嘈杂环境下的稳定表现、复杂指令的精准识别以及低延迟高准确度的响应能力，获得 了众多国际人士的认可。一位来自欧洲的车企代表 ...

Core Viewpoint - The article discusses the competitive landscape of AI companies in Hong Kong, focusing on the IPOs of two firms, Zhiyu (智谱) and MiniMax, which represent different paths in the commercialization of AI technology [3][4]. Group 1: Company Profiles - Zhiyu, led by Chairman Liu Debing, debuted on the Hong Kong Stock Exchange with an IPO price of HKD 116.2, reaching a market capitalization of over HKD 50 billion and raising HKD 4.3 billion, with a staggering 1164 times oversubscription [3][9]. - MiniMax, founded by former SenseTime executive Yan Junjie, is expected to follow closely with a valuation exceeding HKD 50 billion, focusing on consumer applications and global expansion [3][12]. Group 2: Business Strategies - Zhiyu adopts a "B-end and G-end foundational model + localization" approach, emphasizing long-term R&D investments, with R&D spending projected to rise from HKD 84.4 million in 2022 to HKD 2.195 billion in 2024 [9][10]. - MiniMax, on the other hand, follows a "C-end multimodal application + globalization" strategy, targeting both domestic and international markets, with a focus on user-friendly AI applications [11][16]. Group 3: Market Positioning - Zhiyu's revenue structure heavily relies on localized and privatized deployments, which accounted for 95.5% of its revenue in 2022, although it aims to increase its cloud API service revenue, which currently stands at 15.2% [20][22]. - MiniMax's Talkie app has gained traction in overseas markets, with over 70% of its revenue coming from international users, showcasing its successful global strategy [26][27]. Group 4: Competitive Landscape - Both companies face intense competition from established players like OpenAI and domestic giants such as ByteDance and Baidu, which are rapidly advancing in AI capabilities [29][30]. - Despite their current successes, both Zhiyu and MiniMax must navigate a challenging environment where they are squeezed by larger competitors, necessitating innovative strategies to maintain their market positions [28][29].

Core Insights - The article introduces OmniAgent, a proactive perception agent developed by Zhejiang University, West Lake University, and Ant Group, addressing pain points in cross-modal alignment and fine-grained understanding in end-to-end omni-modal models [2][7][19] - OmniAgent employs an innovative "think-act-observe-reflect" closed-loop mechanism, transitioning from passive response to active inquiry, which enhances its performance in audiovisual understanding tasks [10][19] Background and Pain Points - End-to-end omni-modal models face high training costs and challenges in cross-modal feature alignment, leading to subpar performance in fine-grained cross-modal understanding [7] - Fixed workflow-based agents rely on rigid, human-defined processes, lacking the flexibility to autonomously plan and gather information based on questions [7] Methodology - OmniAgent's methodology includes a strategic scheduling of video and audio understanding capabilities within an iterative reflection loop, effectively overcoming cross-modal alignment challenges [8][15] - The agent autonomously decides whether to "listen" or "watch" based on the analysis of the question, utilizing a variety of multimodal tools for efficient information retrieval [15] Performance Results - OmniAgent achieved state-of-the-art (SOTA) results in multiple audiovisual understanding benchmarks, with an accuracy of 82.71% on the Daily-Omni Benchmark, surpassing Gemini 2.5-Flash (72.7%) and Qwen3-Omni-30B (72.08%) by over 10% [13] - In the OmniVideoBench, OmniAgent reached an accuracy of 59.1% in long video understanding tasks, significantly outperforming Qwen3-Omni-30B (38.4%) [13] Future Vision - The design of OmniAgent is highly extensible, allowing for the integration of additional modal tools [19] - OmniAgent is positioned to assist in generating high-quality COTT data for the development of next-generation omni-modal models capable of self-tool invocation [19]