大模型市场，直接来了一个「新物种大分叉」。 全球最大AI模型API聚合平台OpenRouter给出数据： 9号到15号这一周，中国大模型的调用量冲到了 4.12万亿Token，第一次超过美国模型的2.94万亿Token。 调用量前五的模型里，中国直接占了四个：MiniMax的M2.5、月之暗面的KimiK2.5、智谱GLM-5、 DeepSeek V3.2。 市场已经彻底分成两拨人：BAT正在「向内收缩」，把模型塞进搜索、电商、办公这些老业务里；新势 力Kimi、智谱、MiniMax则「向外长手」，靠成百上千个Agent死磕开发者生态。 这是商业底层逻辑的彻底分家，大家都在算一笔名为投资回报率的生死账。 01 先得搞清楚一件事：这4.12万亿Token的调用量，到底是谁在用？ 要光看新闻标题，肯定以为中国老百姓突然爱上跟AI唠嗑了；OpenRouter的用户构成很有意 思，上面美国开发者占了47.17%，中国开发者只有6.01%。 这4.12万亿Token，主要是全球那帮程序员「用脚投票」投出来的，跟咱们在抖音刷到的那些AI聊天没 啥关系。 所以，Token调用量比DAU实在多了，日活用户能被红包砸出来，能被 ...

DeepSeek又找到突破大模型推理瓶颈的新方法了！ 智东西2月27日报道，昨天，DeepSeek发布了一项名为DualPath的全新推理系统方案，直指当前大语言模型在智能体应用场景下遭遇的短板 ——KV缓存存储I/O瓶颈。该方案通过引入双路径加载机制，显著提升系统吞吐量，基本消除了KV缓存的I/O开销。 DualPath的核心创新在于开辟了一条从存储直通解码引擎的新通道。KV缓存不再仅由预填充引擎加载，而是可以加载至解码引擎，再通过计 算网络中的RDMA高效传输至预填充端。这一设计不仅缓解了存储端的压力，还避免了网络拥塞，确保延迟敏感型任务不受干扰。 与全局调度器协同后，DualPath实现了动态平衡两端负载，进一步提升资源利用率。在真实智能体工作负载测试中，DualPath将离线推理吞吐 量提升最高达1.87倍，在线服务吞吐量平均提升1.96倍。 在大规模可扩展性方面，DualPath系统在最多1152张GPU上进行了验证。离线推理从2P4D（2K智能体）扩展到48P96D（48K智能体）实现近 线性扩展，任务完成时间基本保持一致。 值得一提的是，与之前DeepSeek发表的许多研究论文类似，这篇论文的第 ...

（来源：民生加银资讯） 近年来，AI应用正逐步经历从"辅助工具"到"协同伙伴"的关键跃迁，深度重塑产业形态与生活场景。在 技术迭代与政策红利的双重驱动下，这一领域或藏着值得深挖的投资机遇噢~一起来看看吧~ AI应用：或成为驱动社会生产生活向前的"车轮" # 首先，从AI产业链的视角出发，了解下AI应用。AI产业链呈现清晰的三层架构，基础层、中间层与应 用层形成协同联动的价值网络。基础层以算力基础设施、基座模型能力和底层工程能力为核心，通过高 额资本投入构建技术护城河；中间层承担工程化基础设施的角色，将基础层的复杂技术标准化、规模 化，缩短技术从实验室到产业端的转化周期，降低应用层开发门槛；而应用层则是技术落地与商业变现 的核心载体。 简单来说，AI应用就像连接AI技术底座与千行百业现实场景的"连接器"，它一边将抽象的算法与数据能 力转化为适配具体场景的可用工具；一边深入产业与生活的核心痛点，让技术红利真正落地为可感知、 可量化的商业效益与生活体验。如果说基础层是AI产业的"发动机"，中间层是"传动轴"，那么应用层就 像是直接驱动社会生产生活向前的"车轮"，对技术能力的落地精度，以及价值变现的节奏与规模都有着 ...

据媒体2月27日报道，在业界对新一代旗舰模型DeepSeek V4的翘首期盼中，DeepSeek团队却悄然放出了一篇新的学术论 文。 这篇论文由DeepSeek联合北大、清华共同撰写，将研究方向投向了决定大模型实际应用落地的关键一环——推理速度，为 日益复杂的AI智能体，提供一套高效的底层系统解决方案。 具体来说，新论文介绍了一个名为DualPath的创新推理系统，专门针对智能体工作负载下的大模型（LLM）推理性能进行 优化。通过引入"双路径读取KV-Cache（类似记忆缓存）"机制，重新分配存储网络负载，将离线推理吞吐量最高提升 1.87 倍，在线服务的每秒智能体运行数平均提升 1.96 倍。 论文在引言部分提到，大模型正从单轮对话机器人和独立推理模型，快速演进为智能体系统 ——能够自主规划、调用工 具，并通过多轮交互解决实际任务。这种应用范式的转变，推动大模型推理工作负载发生重大变革：从传统的人类-大模型 交互，转向人类-大模型-环境交互，交互轮次可达数十甚至数百轮。 面对传闻，DeepSeek依旧保持其一贯的沉默，目前并未进行任何回应。 此前，DeepSeek被大量用户吐槽风格突变，"变冷淡"，从原本细 ...

Accenture and Mistral AI have entered a multi-year agreement to expand the use of AI in organisations across Europe and globally. The collaboration aims to enable companies to implement secure, large-scale AI systems that comply with regional standards. Accenture will integrate Mistral AI’s models and products, such as Mistral AI Studio, into its own operations and client solutions. The partnership involves joint development of enterprise-level AI solutions intended to address challenges faced by busin ...

Core Insights - The construction of an AI power station by xAI, a company owned by Elon Musk, in Southaven, Mississippi, has led to significant resident protests due to continuous noise pollution, highlighting the conflict between AI computational expansion and community well-being [1][3] Group 1: Project Details - xAI has repurposed an idle power plant and installed 27 temporary gas turbines to supply power to its AI data center, resulting in noise levels comparable to those of an airport runway, causing long-term disturbances for local residents [3] - The company plans to invest over $20 billion and has applied to install 41 permanent turbines, which would represent the largest private investment in the state’s history, promising tax revenue and job creation [3] Group 2: Community Response - Local residents have reported minimal effectiveness from a $7 million sound barrier constructed by the city to mitigate noise [3] - Environmental organizations and the NAACP have accused xAI of operating turbines without proper legal permits, emitting harmful pollutants like formaldehyde, and treating vulnerable communities as "environmental sacrifice zones," with plans to file a lawsuit under the Clean Air Act [3] Group 3: Regulatory Issues - Although the state environmental department has not required permits for the temporary equipment, new federal EPA regulations indicate that such devices should have undergone approval [3] - Public hearings have shown no support for the project from the community, and protests are ongoing, with permanent power station permits expected to be approved as early as next month [3]

When Jeff Bezos said that one breakthrough technology would shape Amazon’s destiny, even Wall Street’s biggest analysts were caught off guard. Fast forward a year and Amazon’s new CEO Andy Jassy described generative AI as a “once-in-a-lifetime” technology that is already being used across Amazon to reinvent customer experiences. At the 8th Future Investment Initiative conference, Elon Musk predicted that by 2040 there would be at least 10 billion humanoid robots, with each priced between $20,000 and $25,000 ...

图源：网络 需要说明的是，"蒸馏"是全球部分AI公司训练自家大模型的常用手段，而Anthropic自己，恰恰是"蒸馏"技术的使用者，甚至存在更激进的数据抓取行为。 值得一提的是，Anthropic如今却成了人工智能领域对中国最不友好的公司之一。2025年9月，Anthropic在官方文件中明确封禁对中资企业的服务。其实指 责中国企业进行数据蒸馏，已经成了美国企业惯用套路。就在2026年2月12日，OpenAI向美国国会提交内部备忘录，明确指控DeepSeek通过复杂的混淆手 段，绕过其安全防护对GPT系列模型实施蒸馏行为。 Anthropic这次直接对中国企业"高调指控"，一个重大背景是进入2026年2月，美国AI概念股出现多轮明显下跌，资本市场对人工智能的未来产生了"颠覆传 统商业模式"的担忧。CNN评论也指出，不少所谓大模型，其实只是精心包装过的搜索引擎。 马斯克嘲讽、资本变脸，中美AI大战升级，中国凭什么破局？ 开工第一天，AI圈就上演"大戏"！美国AI独角兽Anthropic突然发难，指控DeepSeek、月之暗面（Kimi）等3家中国AI企业，靠"蒸馏攻击"抄袭技术。 "蒸馏攻击"是什么？先按下不表 ...

Core Insights - The article discusses a new academic paper released by the DeepSeek team in collaboration with Peking University and Tsinghua University, focusing on inference speed optimization for large language models (LLMs) [1] Group 1: Innovation and Technology - The paper introduces an innovative inference system named DualPath, specifically designed to enhance the inference performance of LLMs under agent workloads [1] - The DualPath system implements a "dual-path reading KV-Cache" mechanism, which reallocates storage network load [1] Group 2: Performance Improvements - The offline inference throughput is reported to have increased by up to 1.87 times [1] - The average number of agent operations per second for online services has improved by 1.96 times [1]

Core Viewpoint - The DeepSeek team has released a new academic paper focusing on optimizing inference speed for large language models (LLMs), which is crucial for the practical application of AI agents [4][5]. Group 1: Research and Innovation - The paper, co-authored with Peking University and Tsinghua University, introduces an innovative inference system called DualPath, designed to enhance the performance of LLMs under agent workloads [4]. - The DualPath system employs a "dual-path reading KV-Cache" mechanism, redistributing storage network load, resulting in an offline inference throughput increase of 1.87 times and an average increase of 1.96 times in the number of agent operations per second for online services [4][5]. Group 2: Industry Context and Expectations - The introduction of DualPath addresses the significant changes in inference workloads as LLMs evolve from simple dialogue systems to complex agent systems capable of multi-turn interactions, which can reach dozens or even hundreds of rounds [4]. - There is a growing expectation for the release of DeepSeek's next flagship model, DeepSeek V4, with various rumors about its launch timeline ranging from early February to March [6]. - Recent leaks suggest that DeepSeek is testing a V4 Lite model, codenamed "Sealion-lite," which supports a context window of 1 million tokens and native multimodal inference [6]. Group 3: Market Reactions and Concerns - Despite the technical advancements presented in the paper, there is a sentiment in the industry that such optimizations are seen as a necessity due to GPU shortages, with some viewing it as "dirty work" rather than innovative [5]. - Concerns have been raised among investment institutions that the release of the new model could lead to significant market volatility, similar to the previous year's model launch [6].