2小时前英伟达拟以超130亿美元收购Hugging Face2026年8月27日，多家媒体援引知情人士消息称，英伟达正与开源 AI 平台 Hugging Face 就一笔估值超过130亿美元的整体收购进行深入谈判，交易若达成将把 Hugging Face 的估值在三年内几乎提升三倍，并可能成为英伟达迄今为止最大规模的并购之一。 目前双方尚未达成正式协议，也均拒绝公开置评，谈判仍处早期阶段、存在较大变数。14 来源谈判概况与潜在交易规模6 来源双方背景：GPU 巨头与“AI 的 GitHub”10 来源潜在影响：开源生态、产业格局与监管关注5 来源谈判不确定性与后续观察9 来源本内容由AI生成

活动榜单专题外部活动妙投投资风向标火热报名中投资风向标直播课（9月）每周二四晚7点，复盘行情、传授分析方法、挖掘投资机会，构建投资体系。妙投波动判断场往期回顾活动已结束寻找市场上的预期差–机器人、大模型与猪周期的预期差推演本期聚焦机器人、大模型、猪周期三个方向的预期差推演——哪些预期太高（需要警惕），哪些预期太低（存在机会）。智库·502线上同行往期回顾报名已结束Agent下车间：吉利、美的、杰克科技的应用场景、卡点与投产比真相吉利集团x美的集团x杰克科技的工厂内Agent落地实践智库·502线下闭门会往期回顾报名已结束工业AI在产线上的真场景与零容错防线舜宇集团、协鑫能源等企业的工业AI场景的真实案例复盘早点生活游玩团往期回顾活动已结束早点生活城市出逃计划——跨年不在酒吧，而在粥粥老家我们即将前往的，是一个用雪花书写浪漫的地方——哈尔滨。

东方小巴黎，夜幕下的哈尔滨，是每个亲历者，无不心动的美。

我想邀约你，和我一起回到我的家乡，和我一起感受这片我爱的土地。 这是一次限定的跨年邀约，同时也是一场难忘的冰雪之旅。虎嗅F&M创新节往期回顾2025虎嗅F&M创新节一次聚焦新商业、新消费与AI原生变革的沉浸式节日化体验智库·AI落地研学营报名已结束AI落地研学营-零售消费班参访研学8-12家AI落地典型企业，与决策者面对面，学习10+最新AI应用落地案例。出海·502线下闭门会往期回顾活动已结束出海·502线下闭门会系列虎嗅「出海•502线下闭门会」面向企业决策层，打造纯粹、高端的行业交流场合，是一次实战经验复盘，更是高价值人脉链接的机会。大鲸AI峰会往期回顾精彩回顾大鲸AI峰会 · 零售消费专场聚焦AI应用落地与零售业的最强碰撞。

Join NowEN在线咨询提交需求AI Search Engine OptimizationBrand Monitoring ServiceMonitor your brand across Google Gemini, ChatGPT, Claude & all major AI models.1Add Brand Information2Analyze Brand Keywords & Competitors3Analysis ReportMonitor by Brand DescriptionMonitor by Website AddressYour Brand NameYour Brand Description0 / 500Your Products/ServicesPlease press Enter to confirm products/services, at least 2 requiredSelect Monitoring AI PlatformSelect Monitoring Time1days7days15daysCustomizableStart AnalysisBrand Score RankingRankBrand NameScoreMeet Enterprise Data Customization NeedsCustomized ReportsAutomatically track your brand through exclusive brand reports. Track key KPIs such as brand mentions, brand coverage, or domain referencesOn-Demand SupportRely on our experienced experts to support you in building artificial intelligence search strategiesEnd-to-End SolutionsGuide optimization of enterprise/product/service pages, community sentiment, and tweet contentContact UsMost Extensive CoverageAnalyze responses from mainstream large models and AI searches at home and abroad, including: Yuanqi, MetaCat, ChatGPTComprehensive Competitive AnalysisIdentify strengths and weaknesses, find gaps to compensate for brand blind spotsGrasp Current Market HotspotsObtain real-time intelligence on all brand-related prompts, keywords, and content sourcesBrand Exclusive TrackingMonitor key metrics at all levels, from the overall brand to individual productsEmpowering the Future, Your AI Solution Knowledge BaseEnglish简体中文繁體中文にほんご© 2026AIBase

AI改变搜索引擎模式超过60%的用户开始使用AI助手查询产品和服务推荐，而不是传统搜索引擎。当用户问“最好的项目管理工具是什么？”时，你的品牌是否出现在回答中？用户提问哪个品牌的智能手表更适合运动？如果AI没有推荐你的品牌，你就失去了这个潜在客户

AI NEWSLatest AI NewsArticleByteDance Seed Adjusts: Comprehensive Restructuring of Pre-training, Reinforcement Learning, and Agent DevelopmentPublished in Latest AI NewsTime :Aug 21, 2026Read :3minuteAccording to media reports, the Seed department of ByteDance’s large model division completed a new round of organizational restructuring last week. Among them, the Seed Foundation Model has newly established four first-level departments, with all department heads reporting to Wu Yonghui. This adjustment further integrates pre-training data, reinforcement learning, and post-training resources, and separates the model capabilities of B-end and C-end products.Pretrain Data is led by Li Chenggang, integrating the previously scattered teams for pre-training data in text models, programming, visual understanding, and speech, and will be responsible for the multimodal data and ultra-large model pre-training data for the new Omni model. Horizon RL is led by Tang Shengyu, integrating the post-training, reasoning, and visual understanding teams, focusing on advancing reinforcement learning to enhance the basic intelligence ceiling of the model.Product Posttrain-Work is led by Qin Yujia, mainly targeting B-end applications, responsible for the integration and release of Agentic models, and optimizing Agentic capabilities for office scenarios, providing support for the task modes of products such as Douba and Dola. The original Application team will continue to be retained and renamed Product Posttrain-Chat, led by Zhu Wenjia, mainly targeting C-end applications, responsible for the integration and release of dialogue models, forming a division of labor with the Work team.This adjustment shows that Seed is further strengthening the model development pipeline from data, pre-training, to reinforcement learning and product post-training, and establishing a differentiated model system for enterprise office agents and consumer dialogue scenarios. ByteDance has not yet responded so far.

AI NEWSLatest AI NewsArticleTaoTie Launches iO AI Glasses, Price 2499 Yuan with 24/7 AI Memory SupportPublished in Latest AI NewsTime :Aug 21, 2026Read :4minuteRayNeo introduced the new iO series at the 2026 RayNeo AI Glasses product launch event. The first product, RayNeo iO, is positioned as a “human enhancement AI glasses” with a price of 2499 yuan, starting at a promotional price of 1996 yuan. It supports prescriptions ranging from 0 to 1000 diopters and is now available for purchase on JD.com, Tmall, Douyin, and offline eyewear channels. According to Counterpoint and IDC data, RayNeo ranked first in China and globally in AR glasses shipments in Q1 2026.RayNeo iO uses a dual-eye display architecture and Blue Lake waveguide technology. The OC area has a 93% light transmission rate, while the non-display area exceeds 98%. The eye illumination is 1800 nits, supporting an equivalent 33-inch display and DC dimming. The device features a magnesium-aluminum alloy front frame and titanium alloy temples, weighing 34 grams. It is equipped with a “Fenghuo Light Engine Nano” smaller than a red bean. The product supports Smart Crown smart dial, nod confirmation, and shake cancellation, and can display information such as weather, schedule, and to-do lists.Regarding AI capabilities, RayNeo iO initially integrates DeepSeek V4Pro and Qwen 3.7Max. Later, it will support Kimi K3 and RayNeo’s self-developed RayNeo AI large model. The battery lasts up to two days. RayNeo’s all-day memory engine can process information such as sound and location, forming a 24-hour context and automatically generating daily summaries. The real-time prompt function can also identify questions in conversations and display reference answers on the lens. The product supports translation into 55 languages and 109 accents, as well as an AI scripter.Privacy-wise, RayNeo iO features a camera-free design. When recording and translation functions are activated, the privacy indicator light turns on simultaneously and cannot be disabled through software. The original conversation data is anonymized and de-identified on the device side. Users can export, migrate, or completely delete the data.

AI NEWSLatest AI NewsArticleAnthropic’s Star Feature Launched! The Computer Operation, Skills, and Files APIs Are Fully AvailablePublished in Latest AI NewsTime :Aug 21, 2026Read :4minuteAnthropic has officially announced that its highly anticipated Computer use, Skills API, and Files API on the Claude Platform are now fully available (Generally Available). This update not only introduces a brand-new Browser use tool but also deeply optimizes underlying capabilities, aiming to empower enterprises to easily build production-grade agents that can truly operate software, apply team expertise, and deliver finished products.Regarding core technology and capability upgrades, the computer use function has achieved significant breakthroughs. The updated tool allows Claude to perform multiple consecutive actions in a single model call, greatly reducing task completion time and the number of calls. Additionally, the computer use tool now meets the standards of the Health Insurance Portability and Accountability Act (HIPAA) and can be used for regulated healthcare workloads under a Business Associate Agreement (BAA).As the absolute highlight of this release, the new Browser use tool is officially launched. This tool inherits the advantages of multi-action rounds and innovatively introduces web structure parsing capabilities. Unlike traditional screenshot-based clicking, it can directly identify and interact with specific input fields or buttons on web pages, enabling more accurate and efficient automation within web applications.In terms of knowledge accumulation and document management ecosystem, the Skills API and Files API provide strong underlying support. The Skills API allows enterprises to upload and manage their own proprietary skill folders, which include industry-standard instructions, scripts, and templates, and run securely within Claude’s code execution sandbox, eliminating the need for enterprises to host them themselves. The Files API provides a stable and efficient storage space for agents to handle documents, supporting the upload of files such as PDFs or spreadsheets and referencing them directly by ID in subsequent requests. It also implements an automatic file expiration mechanism, with a fivefold increase in rate limits, and offers up to 1TB of dedicated storage space per organization.With the full implementation of these major features, developers can continue to use the Skills API and Files API through Microsoft Foundry. The updated computer use and browser use tools will soon be available on Google Cloud’s Vertex AI. Enterprises and developers across industries are accelerating their efforts to leverage this powerful toolchain to comprehensively revolutionize the level of automation in complex business processes.

AI NEWSLatest AI NewsArticleOpenAI Fully Open Sources Codex Harness AI Programming Intelligent Agent, Further Opening Up Its Core CapabilitiesPublished in Latest AI NewsTime :Aug 21, 2026Read :3minuteOpenAI has further opened up the core capabilities of Codex Harness, making the intelligent agent operating framework for different product forms of Codex available to developers. Codex Harness is the core agent runtime environment behind Codex, responsible for connecting the model, users, and tools, and managing processes such as task execution, context, sessions, and code operations. OpenAI previously open-sourced the Codex CLI and related core code, and continues to improve the capabilities of Harness and App Server.From an architectural perspective, Codex Harness is not merely a code generation tool, but a complete agent execution framework. Its core includes Agent Loop, thread lifecycle and persistence, configuration and authentication, tool execution, and extension mechanisms. Through Codex App Server, developers can integrate Harness with different clients using the bidirectional JSON-RPC protocol and obtain real-time event streams, thereby building IDEs, desktop applications, and other customized agent products.Currently, Codex Harness supports various usage scenarios, including Codex Web, CLI, IDE extensions, and macOS applications. OpenAI previously also publicly shared the engineering practice of “code entirely written by Codex” in Harness: an internal product generated about 1 million lines of code within about 5 months and completed approximately 1,500 Pull Requests, with the team’s core philosophy shifting from “manually writing code” to “human setting goals, agent execution.“This further opening of Harness indicates that the competition in AI programming is extending from model capabilities to agent runtime, tool invocation, context management, and engineering feedback loops. As the underlying Harness gradually becomes standardized, developers will be able to build more rich professional AI programming and software engineering workflows on top of Codex’s core agent capabilities.

AI NEWSLatest AI NewsArticlePrice Cut to 1/40! Zhipu Opensources Native Multimodal Large Model GLM-5.3-FlashPublished in Latest AI NewsTime :Aug 27, 2026Read :4minuteRecently, Zhipu officially launched and open-sourced the first native multimodal model of the GLM-5 series, GLM-5.3-Flash (320B-A18B), promoting cutting-edge artificial intelligence technology to be more accessible with leading performance and ultra-low pricing.The model’s capabilities have reached the global forefront. In the Artificial Analysis Intelligence Index evaluation, it scored 57 points, equaling Claude Opus4.8, with comparable programming abilities and superior comprehensive performance compared to the larger parameter model GLM-5.2. Before its official release, the model was tested publicly under the anonymous identity “Ox-Alpha” and topped the usage charts on OpenCode and OpenRouter platforms.The cost advantage is a core highlight. The price of GLM-5.3-Flash is only one-tenth of the same series GLM-5.3, with a limited-time discount as low as one-twentieth, equivalent to one-fortieth the price of Opus4.8, significantly lowering the entry barrier for using cutting-edge models.In terms of architecture, the model adopts a hybrid structure of sparse attention and linear attention, making it the first open-source cutting-edge model to implement this solution, combined with manifold constraint super connection technology. Compared to previous generations, the attention computation has decreased by 3.01 times, and the KV cache has been reduced by 4.44 times, balancing long context capabilities with inference costs. At the same time, it is natively integrated with visual capabilities, capable of independently completing tasks such as front-end development, 3D modeling, and document creation using visual feedback, showing strong performance in professional scenarios like financial reports and legal documents.Notably, the model’s full traffic is supported by domestic chip clusters. The team developed a self-researched inference engine, relying on the EPD separation architecture and multiple memory optimization solutions, achieving a threefold improvement in end-to-end inference performance, with hardware utilization efficiency comparable to mainstream NVIDIA GPUs, proving the feasibility of domestic computing power supporting large-scale cutting-edge model inference.Currently, GLM-5.3-Flash is fully open-sourced, offering API interfaces, online experience channels, and has been launched on intelligent Agent platforms such as ZCode and AutoClaw. It also offers daily experience cards for ten thousand uses, open for global developers to use.Online ExperienceZ.ai: https://chat.z.aiZhipu Qingyan App: https://chatglm.cn

AI NEWSLatest AI NewsArticle23-Year-Old Founder’s AI Assistant Instinct Valued at $2.5 Billion, Raised $350 Million in Six MonthsPublished in Latest AI NewsTime :Aug 27, 2026Read :3minuteInstinct, an AI startup founded by 23-year-old Noah Shinn and only a year old, has surged to an impressive valuation during the AI boom this summer. According to a report by The Wall Street Journal on Wednesday, the company just completed a $250 million Series B funding round, bringing its total funding to $350 million and its valuation up to $2.5 billion. The round was co-led by Index Ventures and Benchmark.An AI agent that helps you “organize your life"Instinct, launched by Spear Street Technology, positions itself as an AI agent that efficiently manages users’ lives. By connecting it to various apps and devices, users can communicate with it via text messages and calls; Shinn excitedly listed early users who used it to plan cross-border road trips, buy groceries, purchase concert tickets, cancel subscriptions worth hundreds of dollars, and even plan weddings on social media.The product is currently in a private beta phase, and its website has a very basic style, but it has already sparked controversy over privacy issues. Users are concerned about the broad permissions it requests and its allegedly invasive terms of service, which were already flagged as a potential problem in a previous TechCrunch report—when an AI agent takes over your life, convenience and loss of control are just a line apart.This funding round is the latest example of the Silicon Valley narrative of “hot AI assistants.” The combination of young founders, viral growth, and top-tier VC backing has allowed Instinct to secure a huge stake before even officially launching; however, the shadow of privacy concerns remains, and whether it can turn this hype into sustainable trust will be the real question behind its $2.5 billion valuation.