[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"portal-settings:stajic:zh":3,"public-menus:all":38,"post:google-io-2026-gemini-omni-and-gemini-3-5:zh":221,"related:post:google-io-2026-gemini-omni-and-gemini-3-5:zh:1":680},{"statusCode":4,"data":5,"message":37},200,{"tenantId":6,"lang":7,"defaultLang":8,"siteUrl":9,"contactEmail":10,"brandName":11,"logoUrl":12,"siteName":11,"siteDescription":13,"ogImage":10,"robotsIndex":14,"socialLinks":10,"reservedSlugs":10,"seoPolicy":15},"stajic","zh","de","https:\u002F\u002Fstajic.de",null,"Stajic Platform","\u002FLogo_Planet.svg","Stajic Portal",true,{"branding":16,"relatedContent":17,"crossDomainLinks":18},{"logoUrl":12},{"enabled":14},[19,22,25,28,31,34],{"url":20,"label":21,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Ffigure.rocks","figure.rocks",{"url":23,"label":24,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Floving.rocks","loving.rocks",{"url":26,"label":27,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.com","bazify.com",{"url":29,"label":30,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.de","bazify.de",{"url":32,"label":33,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.at","bazify.at",{"url":35,"label":36,"isActive":14,"showInFooter":14,"includeInSameAs":14},"https:\u002F\u002Fbazify.ba","bazify.ba","Portal settings resolved",[39,45],{"id":40,"name":41,"location":42,"isActive":14,"isDefault":43,"items":44},1,"main-navigation","header",false,[],{"id":46,"name":47,"location":48,"isActive":14,"isDefault":14,"items":49},4,"main-menu","sidebar",[50,66,79,93,103,118,133],{"id":51,"title":52,"url":60,"target":61,"icon":62,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":64,"portfolioId":10,"children":65},"item-18",{"de":53,"en":54,"es":55,"fr":56,"it":54,"ru":57,"sr":58,"zh":59},"Startseite","Home","Inicio","Accueil","Главная","Почетна","首页","\u002Ffull-stack-web-developer-munich-performance-seo-and-maintainable-builds","_self","i-lucide-home","page",111,[],{"id":67,"title":68,"url":75,"target":61,"icon":76,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":77,"portfolioId":10,"children":78},"item-22",{"de":69,"en":69,"es":70,"fr":69,"it":71,"ru":72,"sr":73,"zh":74},"Vision","Visión","Visione","Видение","Визија","想象","\u002Fueber-uns-webdesign-muenchen-webaplikation","i-lucide-eye",113,[],{"id":80,"title":81,"url":89,"target":61,"icon":90,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":91,"portfolioId":10,"children":92},"item-19",{"de":82,"en":83,"es":84,"fr":83,"it":85,"ru":86,"sr":87,"zh":88},"Leistungen","Services","Servicios","Servizi","Услуги","Услуге","服务","\u002Fservices-dienstleistungen-muenchen","i-lucide-wrench",116,[],{"id":94,"title":95,"url":99,"target":61,"icon":100,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":101,"portfolioId":10,"children":102},"item-23",{"de":96,"en":96,"es":96,"fr":96,"it":96,"ru":97,"sr":97,"zh":98},"Blog","Блог","博客","\u002Fblog","i-lucide-book-open",112,[],{"id":104,"title":105,"url":114,"target":61,"icon":115,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":116,"portfolioId":10,"children":117},"item-32",{"de":106,"en":107,"es":108,"fr":109,"it":110,"ru":111,"sr":112,"zh":113},"Neue Technologien","New Technologies","Nuevas tecnologías","Nouvelles technologies","Nuove tecnologie","Новые технологии","Нове технологије","新技术！","\u002Fneue-webtechnologien","i-lucide-sparkles",122,[],{"id":119,"title":120,"url":129,"target":61,"icon":130,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":131,"portfolioId":10,"children":132},"item-20",{"de":121,"en":122,"es":123,"fr":124,"it":125,"ru":126,"sr":127,"zh":128},"Kontakt","Contact us!","Contacto","Contact","Contatto","Контакт","Контактирајте нас","联系我们！","\u002Fcontact","i-lucide-mail",115,[],{"id":134,"title":135,"url":144,"target":61,"icon":145,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":147},"item-21",{"de":136,"en":137,"es":138,"fr":139,"it":140,"ru":141,"sr":142,"zh":143},"Unsere Arbeit","Our Work","Nuestro trabajo","Nos réalisations","I nostri lavori","Наши работы","Наши радови","文件夹","\u002Fportfolio","i-lucide-briefcase",114,[148,161,175,191,206],{"id":149,"title":150,"url":144,"target":61,"icon":159,"isActive":14,"type":63,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":146,"portfolioId":10,"children":160},"item-24",{"de":151,"en":152,"es":153,"fr":154,"it":155,"ru":156,"sr":157,"zh":158},"Alle Projekte","All Projects","Todos los proyectos","Tous les projets","Tutti i progetti","Все проекты","Сви пројекти","所有项目","i-lucide-grid-3x3",[],{"id":162,"title":163,"url":171,"target":61,"icon":172,"isActive":14,"type":173,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":10,"children":174},"item-29",{"de":164,"en":165,"es":166,"fr":167,"it":168,"ru":169,"sr":170,"zh":143},"Local Roots, Global Reach","Local Roots - Global Reach","Empresa local ","Entreprise locale","Azienda locale","Местная компания","Локално предузеће глобално тржиште","\u002Fportfolio\u002Flocal-roots-global-reach-communication-media-systems-for-modern-business","i-lucide-globe-2","custom",[],{"id":176,"title":177,"url":186,"target":61,"icon":187,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":189,"children":190},"menu-2916d060-8681-482a-af9a-c107a8bc30d8",{"de":178,"en":179,"es":180,"fr":181,"it":182,"ru":183,"sr":184,"zh":185},"Von Rohdaten zu verifizierbarem KI-Kontext: Eine praktische Architektur für agentenbereite Informationen","From Raw Data to Verifiable AI Context:","De datos sin procesar a contexto de IA verificable: una arquitectura práctica para información lista para agentes","Des données brutes au contexte d'IA vérifiable : une architecture pratique pour des informations prêtes pour les agents","Dai dati grezzi a un contesto IA verificabile: Un'architettura pratica per informazioni pronte per gli agenti","От необработанных данных к верифицируемому контексту ИИ: практическая архитектура информации, готовой для агентов","Od sirovih podataka do proverljivog AI konteksta: Praktična arhitektura za informacije spremne za agente","从原始数据到可验证的 AI 上下文：面向智能体就绪信息的实用架构","\u002Fportfolio\u002Ffrom-raw-data-to-verifiable-ai-context-a-practical-architecture-for-agent-ready-information","i-lucide-database","portfolio",22,[],{"id":192,"title":193,"url":202,"target":61,"icon":203,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":204,"children":205},"menu-e2061cd7-ff53-49d5-a0e8-3d49090577c0",{"de":194,"en":195,"es":196,"fr":197,"it":198,"ru":199,"sr":200,"zh":201},"Präsenzplattformen — Unternehmenskommunikation","Presence Platforms — Enterprise Communication","Plataformas de presencia — Comunicación empresarial","Plateformes de présence — Communication d'entreprise","Presence Platforms — Comunicazione aziendale","Платформы присутствия — Корпоративные коммуникации","Presence platforme — Poslovna komunikacija","Presence 平台 — 企业通信","\u002Fportfolio\u002Fpresence-platforms-enterprise-communication-that-wins-without-competing","i-lucide-badge-percent",21,[],{"id":207,"title":208,"url":217,"target":61,"icon":218,"isActive":14,"type":188,"productId":10,"categoryId":10,"shopCategoryId":10,"articleId":10,"pageId":10,"portfolioId":219,"children":220},"menu-a781582e-e90d-418e-b8c9-b070bc37c37e",{"de":209,"en":210,"es":211,"fr":212,"it":213,"ru":214,"sr":215,"zh":216},"Maximierung des POS-ROI durch Konsistenz","Maximizing POS ROI Through Consistent","Maximizar el ROI del POS a través de constante","Maximiser le ROI du PDV grâce à la cohérence","Massimizzare il ROI del POS attraverso costante","Максимизация ROI POS за счет последовательного","Maksimizacija POS ROI-ja kroz dosledno","通过持续最大化 POS 投资回报率","\u002Fportfolio\u002Ffrom-vision-to-value-maximizing-pos-roi-through-consistent-ci-creation-and-strategic-branding","i-lucide-briefcase-business",11,[],{"statusCode":4,"data":222,"message":679},{"id":223,"title":224,"slug":225,"content":226,"contentJson":227,"excerpt":395,"featuredImage":396,"featuredImageAlt":397,"featuredImageCaption":10,"featuredImageTitle":10,"featuredImageCopyright":10,"featuredImageAuthor":10,"featuredImageSourceUrl":10,"featuredImageLicense":10,"featuredImageIsAiGenerated":43,"status":398,"publishedAt":399,"createdAt":400,"updatedAt":401,"seoLocalePaths":402,"categories":411,"author":412,"translations":417},"450","Google I\u002FO 2026：Gemini Omni、Gemini 3.5 以及驱动自主式AI的计算层","google-io-2026-gemini-omni-and-gemini-3-5","\u003Cp>\u003Cb>Google I\u002FO 2026\u003C\u002Fb> 将谷歌AI战略的计算层重新置于中心。主题演讲充满了智能体、开发者工具、可穿戴界面和产品集成，但所有这些层都依赖于两个模型方向：\u003Cb>Gemini Omni\u003C\u002Fb> 和 \u003Cb>Gemini 3.5\u003C\u002Fb>。一个推动多模态创作和编辑。另一个通过行动推动前沿智能，尤其是通过 Gemini 3.5 Flash。\u003C\u002Fp>\n\u003Cp>本文是更大的 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026 架构中心\u003C\u002Fa> 中的\u003Cb>计算层分支\u003C\u002Fb>。开发者工具方面在 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Antigravity、AI Studio 和 Google DevTools\u003C\u002Fa> 中介绍。设备表面故事在 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Android XR 和智能眼镜\u003C\u002Fa> 中介绍。产品表面故事在 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">搜索、工作空间和购物中的智能产品\u003C\u002Fa> 中介绍。\u003C\u002Fp>\n\u003Cp>重要的是要避免将 Gemini Omni 和 Gemini 3.5 解读为一个通用的“新AI模型”公告。它们解决不同的产品和基础设施问题。Omni 是关于任意输入的创意生成，从视频开始。Gemini 3.5 是关于面向智能体、编码和长周期工作流的快速、行动导向的智能。这种分离对开发者、产品团队和平台架构师来说很重要。\u003C\u002Fp>\n\u003Ch2>主题演讲背后的计算层\u003C\u002Fh2>\n\u003Cp>谷歌自己的 I\u002FO 2026 集合将活动围绕两个新模型展开：Gemini Omni 和 Gemini 3.5。Gemini Omni 被描述为在世界理解、多模态和编辑方面的飞跃，能够从文本、图像、音频和视频的输入创建输出，从视频开始。Gemini 3.5 Flash 被介绍为 Gemini 3.5 系列中的第一个模型，结合了前沿智能与行动。\u003C\u002Fp>\n\u003Cul>\u003Cli>Gemini Omni 专注于多模态创作和编辑，从视频开始。\u003C\u002Fli>\u003Cli>Gemini 3.5 Flash 专注于面向智能体、编码和复杂工作流的行动级速度。\u003C\u002Fli>\u003Cli>模型分离反映了谷歌更广泛的平台分离：创建更丰富的媒体，然后在产品和工具中更快地行动。\u003C\u002Fli>\u003Cli>当这些模型连接到 Antigravity、AI Studio、搜索、工作空间、Android XR 和 Gemini 驱动的设备时，真正的战略价值才会显现。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">模型故事不仅仅是更大的上下文或更好的演示。它是角色专业化：一边是创作，另一边是行动。\u003Ccite class=\"block mt-2 text-sm\">— 计算层解读\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Ch2>Gemini Omni：从任意输入进行创作\u003C\u002Fh2>\n\u003Cp>Gemini Omni 是谷歌将推理与创作融合的最直接尝试。官方定位很明确：它可以从任何输入创建任何内容，从视频开始。该模型接受文本、图像、音频和视频的组合，然后生成基于 Gemini 现实世界知识的高质量视频。它还支持对话式视频编辑，这在实际创意工作流中可能比一次性生成更重要。\u003C\u002Fp>\n\u003Cp>最后一点很重要。创意AI的实际价值很少仅仅是“生成一个视频”。生产工作流需要修改、替换、时机、风格一致性以及围绕变化的对话。如果 Omni 能在保持上下文的同时使编辑更具对话性，它就会超越新奇生成器，成为一种创意界面。\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>\u002F\u002F Gemini Omni 应被视为创意工作流模型\ninput = { text: "在雨中的赛博朋克城市展示产品发布", image: "品牌参考.png", audio: "旁白.wav", video: "粗剪.mp4"\n} output = GeminiOmni.generateVideo(input)\nrevision = GeminiOmni.edit(output, "让光线更温暖，并保持相同的产品设计")\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>对于构建媒体工具、电商产品视觉、教育内容、游戏原型、广告工作流或社交视频管道的团队来说，这是 I\u002FO 2026 中最相关的部分。界面从手动时间线编辑转向基于意图的多模态编辑。这并不会消除创意判断，但会改变创意劳动的投入点。\u003C\u002Fp>\n\u003Ch2>Gemini 3.5 Flash：具有行动能力的前沿智能\u003C\u002Fh2>\n\u003Cp>Gemini 3.5 Flash 是计算故事的另一半。谷歌将 Gemini 3.5 系列描述为结合前沿智能与行动，并将 3.5 Flash 定位为该系列的第一个版本。重点在于智能体、编码、复杂的长期任务和现实世界的实用性。这种措辞并非偶然。它直接指向能够执行操作的系统，而不仅仅是回答问题的系统。\u003C\u002Fp>\n\u003Cp>对于开发者来说，“Flash”之所以重要，是因为智能体工作流对延迟敏感。一个聪明但缓慢的模型可以用于研究或仔细写作。但它很难融入终端原生自动化、IDE循环、多步骤工具使用、后台智能体以及需要即时反馈的产品体验中。谷歌显然试图让快速的前沿行为成为智能体工作流的默认引擎。\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>\u002F\u002F 经典模型用法\nanswer = model.generate("解释这个函数") \u002F\u002F 智能体模型用法\nplan = model.plan(task)\ntoolResult = tools.run(plan.nextStep)\nverification = model.verify(toolResult) if verification.failed: rollback()\nelse: continueExecution()\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>这种转变就是为什么 Gemini 3.5 与 Antigravity 属于同一集群。Antigravity 需要一个能够规划、执行、验证并持续前进而不将开发环境变成等待室的模型。这就是计算性能成为开发者体验的地方。\u003C\u002Fp>\n\u003Ch2>为什么速度和行动应该在一起\u003C\u002Fh2>\n\u003Cp>谷歌主题演讲中关于速度的表述并非仅仅是营销手段。智能体系统会成倍增加延迟。一个简单的答案只需要一次模型往返。而一个多智能体工作流可能需要多次：规划、工具选择、代码编辑、终端执行、验证、修复、总结和人工审核。如果每一步都很慢，整个系统将变得不可用。\u003C\u002Fp>\n\u003Cul>\u003Cli>快速输出能改善交互式编码循环。\u003C\u002Fli>\u003Cli>低延迟让多步骤智能体感觉不那么脆弱。\u003C\u002Fli>\u003Cli>当智能体在后台维护任务中运行时，高吞吐量很重要。\u003C\u002Fli>\u003Cli>稳定的推理很重要，因为更快的错误操作仍然是错误操作。\u003C\u002Fli>\u003Cli>验证仍然至关重要，因为速度本身并不等于信任。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">有用的智能体不是只回答一次的模型。而是能够持续行动，而不会让工作流感觉像排队一样的模型。\u003Ccite class=\"block mt-2 text-sm\">— 智能体运行时视角\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Ch2>Gemini Omni 与 Gemini 3.5：不同任务，相同堆栈\u003C\u002Fh2>\n\u003Cp>理解 I\u002FO 2026 计算公告最清晰的方式是区分任务。Gemini Omni 是一个多模态创作和编辑模型。Gemini 3.5 Flash 是一个面向智能体和编码的行动导向模型。它们是互补的，不可互换。一个产品团队可能使用 Omni 生成媒体，并使用 Gemini 3.5 Flash 来编排围绕该媒体的工作流。\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>\u002F\u002F 产品工作流示例\ncreativeAsset = GeminiOmni.generateVideo(brief) agentPlan = Gemini35Flash.plan({ task: "Publish launch campaign", asset: creativeAsset, channels: ["YouTube Shorts", "landing page", "email"]\n}) Gemini35Flash.execute(agentPlan, tools)\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>这就是谷歌所指的更广泛架构。创意模型生成更丰富的工件。行动模型协调围绕这些工件的任务。产品界面将这些能力暴露给用户。开发者工具提供编排和验证。设备界面使助手在更多场景中可用。\u003C\u002Fp>\n\u003Ch2>这对开发者意味着什么\u003C\u002Fh2>\n\u003Cp>开发者不应将 Gemini 3.5 Flash 仅仅视为另一个模型选项。它改变了可以在谷歌堆栈之上构建的应用程序形态。如果一个模型能够处理更长的任务序列、可靠地使用工具，并且运行速度快到足以支持真实工作流，那么应用程序就可以从孤立的 AI 功能转变为模型中介的过程。\u003C\u002Fp>\n\u003Cul>\u003Cli>代码智能体可以重构、测试和验证更大的任务单元。\u003C\u002Fli>\u003Cli>产品智能体可以帮助用户在单个流程中从意图转向行动。\u003C\u002Fli>\u003Cli>媒体应用可以将 Omni 生成与 Gemini 3.5 编排相结合。\u003C\u002Fli>\u003Cli>搜索和 Workspace 功能可以从回答问题转向执行任务。\u003C\u002Fli>\u003Cli>移动和 XR 界面可以将 Gemini 用作实时上下文层，而不是遥远的云助手。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cp>风险在于在运营模型成熟之前过度构建。开发者仍然需要版本控制、评估、日志记录、成本控制、回退行为以及针对高影响操作的人工确认。更强的模型并不能消除对工程纪律的需求。它反而增加了这种纪律的价值。\u003C\u002Fp>\n\u003Ch2>这对产品团队意味着什么\u003C\u002Fh2>\n\u003Cp>产品团队应将 Gemini Omni 和 Gemini 3.5 解读为一个信号：谷歌希望 AI 成为整个生态系统的执行层。这既带来了机遇，也带来了风险。机遇是从用户意图到最终输出的路径更快。风险在于，如果智能体决策隐藏在精美的界面背后，产品行为可能变得不那么透明。\u003C\u002Fp>\n\u003Cpre class=\"code-block\">\u003Ccode>agenticProductChecklist = { userIntent: "clear", modelRole: "defined", tools: "permissioned", outputs: "reviewable", failures: "recoverable", costs: "measured", userControl: "explicit"\n}\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>如果一个产品团队无法回答模型被允许做什么、如何评估以及用户如何从错误操作中恢复，那么该集成尚未准备好投入生产。当 Gemini 连接到搜索、Workspace、购物、Android 或可穿戴界面时，这一点尤其重要。\u003C\u002Fp>\n\u003Ch2>现实检验：基准测试并非产品\u003C\u002Fh2>\n\u003Cp>每次模型发布的风险都是基准测试中毒。更好的性能很重要，但生产价值取决于集成质量。一个模型可能更快更强，但由于提示词薄弱、工具策略松散、评估肤浅或回滚路径缺失，仍然可能在特定工作流中失败。\u003C\u002Fp>\n\u003Cul>\u003Cli>创意模型需要一致性、可编辑性和输出控制。\u003C\u002Fli>\u003Cli>行动模型需要工具纪律、验证和安全的失败模式。\u003C\u002Fli>\u003Cli>开发者模型需要在真实代码仓库中具有可重复的行为。\u003C\u002Fli>\u003Cli>产品模型需要用户信任和清晰的控制边界。\u003C\u002Fli>\u003Cli>平台模型需要版本控制、监控和成本可观测性。\u003C\u002Fli>\u003C\u002Ful>\n\u003Cblockquote class=\"border-l-4 border-gray-300 pl-4 italic\">发布模型吸引关注。运营模型决定它能否在生产中存活。\u003Ccite class=\"block mt-2 text-sm\">— 工程现实检验\u003C\u002Fcite>\u003C\u002Fblockquote>\n\u003Ch2>这与 Google I\u002FO 2026 集群的契合点\u003C\u002Fh2>\n\u003Cp>本文涵盖集群的模型层。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026 主枢纽\u003C\u002Fa>解释了完整的架构转变。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Antigravity、AI Studio 和 Google DevTools\u003C\u002Fa>解释了模型如何进入开发者工作流。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Android XR 和智能眼镜\u003C\u002Fa>解释了相同的智能如何进入可穿戴设备表面。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">跨搜索、工作空间和购物的智能体产品\u003C\u002Fa>解释了当这些能力到达消费者和生产力产品时会发生什么。\u003C\u002Fp>\n\u003Ch2>本集群相关文章\u003C\u002Fh2>\n\u003Cul>\u003Cli>\u003Cb>主枢纽：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026：架构转型、智能体 AI 与统一生态系统现实检验\u003C\u002Fa>\u003C\u002Fli>\u003Cli>\u003Cb>开发者工具：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Google I\u002FO 2026：Antigravity、AI Studio 和 Google DevTools\u003C\u002Fa>\u003C\u002Fli>\u003Cli>\u003Cb>Android、XR 和设备表面：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Google I\u002FO 2026：Android XR 和智能眼镜\u003C\u002Fa>\u003C\u002Fli>\u003Cli>\u003Cb>智能体消费者产品：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">Google I\u002FO 2026：跨搜索、工作空间和购物的智能体产品\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Ful>\n\u003Ch2>最终视角\u003C\u002Fh2>\n\u003Cp>Gemini Omni 和 Gemini 3.5 定义了 Google I\u002FO 2026 背后的计算分工。Omni 是创作层：多模态输入、视频输出、对话式编辑和基于世界的生成。Gemini 3.5 Flash 是行动层：为智能体、编码和长期任务提供快速前沿智能。它们共同解释了为什么主题演讲的其他部分呈现出那样的面貌。Antigravity 需要行动级模型。搜索和工作空间需要能够推理和执行的智能体。Android XR 需要实时上下文理解。主题演讲不仅仅是一个模型故事，但没有这个计算层，其余的故事就无法成立。\u003C\u002Fp>\n\u003Cdiv class=\"ce-delimiter cdx-block my-8\">\u003C\u002Fdiv>",{"time":228,"blocks":229,"version":394},1779228480767,[230,234,237,240,244,247,256,261,264,267,270,274,277,280,283,286,289,292,295,298,306,310,313,316,319,322,325,328,336,339,342,345,348,351,354,357,365,369,372,375,378,385,388,391],{"data":231,"type":233},{"text":232},"\u003Cb>Google I\u002FO 2026\u003C\u002Fb> 将谷歌AI战略的计算层重新置于中心。主题演讲充满了智能体、开发者工具、可穿戴界面和产品集成，但所有这些层都依赖于两个模型方向：\u003Cb>Gemini Omni\u003C\u002Fb> 和 \u003Cb>Gemini 3.5\u003C\u002Fb>。一个推动多模态创作和编辑。另一个通过行动推动前沿智能，尤其是通过 Gemini 3.5 Flash。","paragraph",{"data":235,"type":233},{"text":236},"本文是更大的 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026 架构中心\u003C\u002Fa> 中的\u003Cb>计算层分支\u003C\u002Fb>。开发者工具方面在 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Antigravity、AI Studio 和 Google DevTools\u003C\u002Fa> 中介绍。设备表面故事在 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Android XR 和智能眼镜\u003C\u002Fa> 中介绍。产品表面故事在 \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">搜索、工作空间和购物中的智能产品\u003C\u002Fa> 中介绍。",{"data":238,"type":233},{"text":239},"重要的是要避免将 Gemini Omni 和 Gemini 3.5 解读为一个通用的“新AI模型”公告。它们解决不同的产品和基础设施问题。Omni 是关于任意输入的创意生成，从视频开始。Gemini 3.5 是关于面向智能体、编码和长周期工作流的快速、行动导向的智能。这种分离对开发者、产品团队和平台架构师来说很重要。",{"data":241,"type":42},{"text":242,"level":243},"主题演讲背后的计算层",2,{"data":245,"type":233},{"text":246},"谷歌自己的 I\u002FO 2026 集合将活动围绕两个新模型展开：Gemini Omni 和 Gemini 3.5。Gemini Omni 被描述为在世界理解、多模态和编辑方面的飞跃，能够从文本、图像、音频和视频的输入创建输出，从视频开始。Gemini 3.5 Flash 被介绍为 Gemini 3.5 系列中的第一个模型，结合了前沿智能与行动。",{"data":248,"type":255},{"items":249,"style":254},[250,251,252,253],"Gemini Omni 专注于多模态创作和编辑，从视频开始。","Gemini 3.5 Flash 专注于面向智能体、编码和复杂工作流的行动级速度。","模型分离反映了谷歌更广泛的平台分离：创建更丰富的媒体，然后在产品和工具中更快地行动。","当这些模型连接到 Antigravity、AI Studio、搜索、工作空间、Android XR 和 Gemini 驱动的设备时，真正的战略价值才会显现。","unordered","list",{"data":257,"type":260},{"text":258,"caption":259},"模型故事不仅仅是更大的上下文或更好的演示。它是角色专业化：一边是创作，另一边是行动。","计算层解读","quote",{"data":262,"type":42},{"text":263,"level":243},"Gemini Omni：从任意输入进行创作",{"data":265,"type":233},{"text":266},"Gemini Omni 是谷歌将推理与创作融合的最直接尝试。官方定位很明确：它可以从任何输入创建任何内容，从视频开始。该模型接受文本、图像、音频和视频的组合，然后生成基于 Gemini 现实世界知识的高质量视频。它还支持对话式视频编辑，这在实际创意工作流中可能比一次性生成更重要。",{"data":268,"type":233},{"text":269},"最后一点很重要。创意AI的实际价值很少仅仅是“生成一个视频”。生产工作流需要修改、替换、时机、风格一致性以及围绕变化的对话。如果 Omni 能在保持上下文的同时使编辑更具对话性，它就会超越新奇生成器，成为一种创意界面。",{"data":271,"type":273},{"code":272},"\u002F\u002F Gemini Omni 应被视为创意工作流模型\ninput = { text: \"在雨中的赛博朋克城市展示产品发布\", image: \"品牌参考.png\", audio: \"旁白.wav\", video: \"粗剪.mp4\"\n} output = GeminiOmni.generateVideo(input)\nrevision = GeminiOmni.edit(output, \"让光线更温暖，并保持相同的产品设计\")","code",{"data":275,"type":233},{"text":276},"对于构建媒体工具、电商产品视觉、教育内容、游戏原型、广告工作流或社交视频管道的团队来说，这是 I\u002FO 2026 中最相关的部分。界面从手动时间线编辑转向基于意图的多模态编辑。这并不会消除创意判断，但会改变创意劳动的投入点。",{"data":278,"type":42},{"text":279,"level":243},"Gemini 3.5 Flash：具有行动能力的前沿智能",{"data":281,"type":233},{"text":282},"Gemini 3.5 Flash 是计算故事的另一半。谷歌将 Gemini 3.5 系列描述为结合前沿智能与行动，并将 3.5 Flash 定位为该系列的第一个版本。重点在于智能体、编码、复杂的长期任务和现实世界的实用性。这种措辞并非偶然。它直接指向能够执行操作的系统，而不仅仅是回答问题的系统。",{"data":284,"type":233},{"text":285},"对于开发者来说，“Flash”之所以重要，是因为智能体工作流对延迟敏感。一个聪明但缓慢的模型可以用于研究或仔细写作。但它很难融入终端原生自动化、IDE循环、多步骤工具使用、后台智能体以及需要即时反馈的产品体验中。谷歌显然试图让快速的前沿行为成为智能体工作流的默认引擎。",{"data":287,"type":273},{"code":288},"\u002F\u002F 经典模型用法\nanswer = model.generate(\"解释这个函数\") \u002F\u002F 智能体模型用法\nplan = model.plan(task)\ntoolResult = tools.run(plan.nextStep)\nverification = model.verify(toolResult) if verification.failed: rollback()\nelse: continueExecution()",{"data":290,"type":233},{"text":291},"这种转变就是为什么 Gemini 3.5 与 Antigravity 属于同一集群。Antigravity 需要一个能够规划、执行、验证并持续前进而不将开发环境变成等待室的模型。这就是计算性能成为开发者体验的地方。",{"data":293,"type":42},{"text":294,"level":243},"为什么速度和行动应该在一起",{"data":296,"type":233},{"text":297},"谷歌主题演讲中关于速度的表述并非仅仅是营销手段。智能体系统会成倍增加延迟。一个简单的答案只需要一次模型往返。而一个多智能体工作流可能需要多次：规划、工具选择、代码编辑、终端执行、验证、修复、总结和人工审核。如果每一步都很慢，整个系统将变得不可用。",{"data":299,"type":255},{"items":300,"style":254},[301,302,303,304,305],"快速输出能改善交互式编码循环。","低延迟让多步骤智能体感觉不那么脆弱。","当智能体在后台维护任务中运行时，高吞吐量很重要。","稳定的推理很重要，因为更快的错误操作仍然是错误操作。","验证仍然至关重要，因为速度本身并不等于信任。",{"data":307,"type":260},{"text":308,"caption":309},"有用的智能体不是只回答一次的模型。而是能够持续行动，而不会让工作流感觉像排队一样的模型。","智能体运行时视角",{"data":311,"type":42},{"text":312,"level":243},"Gemini Omni 与 Gemini 3.5：不同任务，相同堆栈",{"data":314,"type":233},{"text":315},"理解 I\u002FO 2026 计算公告最清晰的方式是区分任务。Gemini Omni 是一个多模态创作和编辑模型。Gemini 3.5 Flash 是一个面向智能体和编码的行动导向模型。它们是互补的，不可互换。一个产品团队可能使用 Omni 生成媒体，并使用 Gemini 3.5 Flash 来编排围绕该媒体的工作流。",{"data":317,"type":273},{"code":318},"\u002F\u002F 产品工作流示例\ncreativeAsset = GeminiOmni.generateVideo(brief) agentPlan = Gemini35Flash.plan({ task: \"Publish launch campaign\", asset: creativeAsset, channels: [\"YouTube Shorts\", \"landing page\", \"email\"]\n}) Gemini35Flash.execute(agentPlan, tools)",{"data":320,"type":233},{"text":321},"这就是谷歌所指的更广泛架构。创意模型生成更丰富的工件。行动模型协调围绕这些工件的任务。产品界面将这些能力暴露给用户。开发者工具提供编排和验证。设备界面使助手在更多场景中可用。",{"data":323,"type":42},{"text":324,"level":243},"这对开发者意味着什么",{"data":326,"type":233},{"text":327},"开发者不应将 Gemini 3.5 Flash 仅仅视为另一个模型选项。它改变了可以在谷歌堆栈之上构建的应用程序形态。如果一个模型能够处理更长的任务序列、可靠地使用工具，并且运行速度快到足以支持真实工作流，那么应用程序就可以从孤立的 AI 功能转变为模型中介的过程。",{"data":329,"type":255},{"items":330,"style":254},[331,332,333,334,335],"代码智能体可以重构、测试和验证更大的任务单元。","产品智能体可以帮助用户在单个流程中从意图转向行动。","媒体应用可以将 Omni 生成与 Gemini 3.5 编排相结合。","搜索和 Workspace 功能可以从回答问题转向执行任务。","移动和 XR 界面可以将 Gemini 用作实时上下文层，而不是遥远的云助手。",{"data":337,"type":233},{"text":338},"风险在于在运营模型成熟之前过度构建。开发者仍然需要版本控制、评估、日志记录、成本控制、回退行为以及针对高影响操作的人工确认。更强的模型并不能消除对工程纪律的需求。它反而增加了这种纪律的价值。",{"data":340,"type":42},{"text":341,"level":243},"这对产品团队意味着什么",{"data":343,"type":233},{"text":344},"产品团队应将 Gemini Omni 和 Gemini 3.5 解读为一个信号：谷歌希望 AI 成为整个生态系统的执行层。这既带来了机遇，也带来了风险。机遇是从用户意图到最终输出的路径更快。风险在于，如果智能体决策隐藏在精美的界面背后，产品行为可能变得不那么透明。",{"data":346,"type":273},{"code":347},"agenticProductChecklist = { userIntent: \"clear\", modelRole: \"defined\", tools: \"permissioned\", outputs: \"reviewable\", failures: \"recoverable\", costs: \"measured\", userControl: \"explicit\"\n}",{"data":349,"type":233},{"text":350},"如果一个产品团队无法回答模型被允许做什么、如何评估以及用户如何从错误操作中恢复，那么该集成尚未准备好投入生产。当 Gemini 连接到搜索、Workspace、购物、Android 或可穿戴界面时，这一点尤其重要。",{"data":352,"type":42},{"text":353,"level":243},"现实检验：基准测试并非产品",{"data":355,"type":233},{"text":356},"每次模型发布的风险都是基准测试中毒。更好的性能很重要，但生产价值取决于集成质量。一个模型可能更快更强，但由于提示词薄弱、工具策略松散、评估肤浅或回滚路径缺失，仍然可能在特定工作流中失败。",{"data":358,"type":255},{"items":359,"style":254},[360,361,362,363,364],"创意模型需要一致性、可编辑性和输出控制。","行动模型需要工具纪律、验证和安全的失败模式。","开发者模型需要在真实代码仓库中具有可重复的行为。","产品模型需要用户信任和清晰的控制边界。","平台模型需要版本控制、监控和成本可观测性。",{"data":366,"type":260},{"text":367,"caption":368},"发布模型吸引关注。运营模型决定它能否在生产中存活。","工程现实检验",{"data":370,"type":42},{"text":371,"level":243},"这与 Google I\u002FO 2026 集群的契合点",{"data":373,"type":233},{"text":374},"本文涵盖集群的模型层。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026 主枢纽\u003C\u002Fa>解释了完整的架构转变。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Antigravity、AI Studio 和 Google DevTools\u003C\u002Fa>解释了模型如何进入开发者工作流。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Android XR 和智能眼镜\u003C\u002Fa>解释了相同的智能如何进入可穿戴设备表面。\u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">跨搜索、工作空间和购物的智能体产品\u003C\u002Fa>解释了当这些能力到达消费者和生产力产品时会发生什么。",{"data":376,"type":42},{"text":377,"level":243},"本集群相关文章",{"data":379,"type":255},{"items":380,"style":254},[381,382,383,384],"\u003Cb>主枢纽：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026：架构转型、智能体 AI 与统一生态系统现实检验\u003C\u002Fa>","\u003Cb>开发者工具：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Google I\u002FO 2026：Antigravity、AI Studio 和 Google DevTools\u003C\u002Fa>","\u003Cb>Android、XR 和设备表面：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Google I\u002FO 2026：Android XR 和智能眼镜\u003C\u002Fa>","\u003Cb>智能体消费者产品：\u003C\u002Fb> \u003Ca href=\"\u002Fzh\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">Google I\u002FO 2026：跨搜索、工作空间和购物的智能体产品\u003C\u002Fa>",{"data":386,"type":42},{"text":387,"level":243},"最终视角",{"data":389,"type":233},{"text":390},"Gemini Omni 和 Gemini 3.5 定义了 Google I\u002FO 2026 背后的计算分工。Omni 是创作层：多模态输入、视频输出、对话式编辑和基于世界的生成。Gemini 3.5 Flash 是行动层：为智能体、编码和长期任务提供快速前沿智能。它们共同解释了为什么主题演讲的其他部分呈现出那样的面貌。Antigravity 需要行动级模型。搜索和工作空间需要能够推理和执行的智能体。Android XR 需要实时上下文理解。主题演讲不仅仅是一个模型故事，但没有这个计算层，其余的故事就无法成立。",{"data":392,"type":393},{},"delimiter","2.31","Google I\u002FO 2026 将 Gemini Omni 和 Gemini 3.5 置于谷歌代理型 AI 战略的核心。本文解析了多模态创作与行动级智能之间的区别，阐释了 Gemini 3.5 Flash 对代理和编码的重要性，以及这些模型如何驱动更广泛的 Google I\u002FO 2026 平台转型。","\u002Fuploads\u002F2026\u002F05\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5-1779227791401-6hzln1.webp","google-io-2026-gemini-omni-and-gemini-3-5-1779227791401-6hzln1","PUBLISHED","2026-05-21T11:01:00.000Z","2026-05-19T21:24:25.091Z","2026-05-21T11:07:38.647Z",{"en":403,"de":404,"sr":405,"es":406,"fr":407,"it":408,"ru":409,"zh":410},"\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Fde\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Fsr\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Fes\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Ffr\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Fit\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Fru\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5","\u002Fzh\u002Fblog\u002Fgoogle-io-2026-gemini-omni-and-gemini-3-5",[],{"id":413,"login":414,"email":415,"displayName":416},"20","rooth8233","","Aleksandar Stajić",[418,583],{"lang":419,"title":420,"content":421,"contentJson":422,"excerpt":582},"en","Google I\u002FO 2026: Gemini Omni, Gemini 3.5, and the Compute Layer Behind Agentic AI","{\"time\":1774828800000,\"blocks\":[{\"data\":{\"text\":\"\u003Cb>Google I\u002FO 2026\u003C\u002Fb> put the compute layer of Google’s AI strategy back in the center. The keynote was full of agents, developer tools, wearable interfaces, and product integrations, but all of those layers depend on two model directions: \u003Cb>Gemini Omni\u003C\u002Fb> and \u003Cb>Gemini 3.5\u003C\u002Fb>. One pushes multimodal creation and editing. The other pushes frontier intelligence with action, especially through Gemini 3.5 Flash.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"This article is the \u003Cb>compute-layer spoke\u003C\u002Fb> inside the larger \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\\\">Google I\u002FO 2026 architectural hub\u003C\u002Fa>. The developer-tooling side is covered in \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\\\">Antigravity, AI Studio, and Google DevTools\u003C\u002Fa>. The device-surface story is covered in \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\\\">Android XR and Intelligent Eyewear\u003C\u002Fa>. The product-surface story is covered in \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\\\">Agentic Products Across Search, Workspace, and Shopping\u003C\u002Fa>.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"The important thing is to avoid reading Gemini Omni and Gemini 3.5 as one generic “new AI model” announcement. They solve different product and infrastructure problems. Omni is about any-input creative generation, starting with video. Gemini 3.5 is about fast, action-oriented intelligence for agents, coding, and long-horizon workflows. That separation matters for developers, product teams, and platform architects.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"The Compute Layer Behind the Keynote\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Google’s own I\u002FO 2026 collection frames the event around two new models: Gemini Omni and Gemini 3.5. Gemini Omni is described as a leap forward in world understanding, multimodality, and editing, able to create output from input across text, images, audio, and video, starting with video. Gemini 3.5 Flash is introduced as the first model in the Gemini 3.5 family, combining frontier intelligence with action.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"Gemini Omni focuses on multimodal creation and editing, starting with video.\",\"Gemini 3.5 Flash focuses on action-grade speed for agents, coding, and complex workflows.\",\"The model split mirrors Google’s broader platform split: create richer media, then act faster across products and tools.\",\"The real strategic value comes when these models are connected to Antigravity, AI Studio, Search, Workspace, Android XR, and Gemini-powered devices.\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"The model story is not just bigger context or better demos. It is role specialization: creation on one side, action on the other.\",\"caption\":\"Compute-layer reading\"},\"type\":\"quote\"},{\"data\":{\"text\":\"Gemini Omni: Creation From Any Input\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Gemini Omni is Google’s most direct attempt to fuse reasoning with creation. The official positioning is clear: it can create anything from any input, starting with video. The model accepts combinations of text, image, audio, and video, then generates high-quality video grounded in Gemini’s real-world knowledge. It also supports conversational video editing, which is likely to matter more than one-shot generation in actual creative workflows.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"That last point is important. The practical value of creative AI is rarely only “generate a video.” Production workflows require revision, replacement, timing, style continuity, and conversation around changes. If Omni can make editing more conversational while preserving context, it becomes more than a novelty generator. It becomes a creative interface.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"\u002F\u002F Gemini Omni should be read as a creative workflow model\\ninput = { text: \\\"Show a product launch in a rainy cyberpunk city\\\", image: \\\"brand-reference.png\\\", audio: \\\"voiceover.wav\\\", video: \\\"rough-cut.mp4\\\"\\n} output = GeminiOmni.generateVideo(input)\\nrevision = GeminiOmni.edit(output, \\\"Make the lighting warmer and keep the same product design\\\")\"},\"type\":\"code\"},{\"data\":{\"text\":\"For teams building media tools, ecommerce product visuals, education content, game prototyping, advertising workflows, or social video pipelines, this is the part of I\u002FO 2026 that matters most. The interface moves from manual timeline editing toward intent-based multimodal editing. That does not remove creative judgment, but it changes where creative labor is spent.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"Gemini 3.5 Flash: Frontier Intelligence With Action\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Gemini 3.5 Flash is the other half of the compute story. Google describes the Gemini 3.5 family as combining frontier intelligence with action, and positions 3.5 Flash as the first release in that family. The emphasis is on agents, coding, complex long-horizon tasks, and real-world utility. That language is not accidental. It points directly toward systems that do things, not just systems that answer.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"For developers, “Flash” matters because agentic workflows are latency-sensitive. A model that is smart but slow can work for research or careful writing. It is much harder to fit into terminal-native automation, IDE loops, multi-step tool usage, background agents, and product experiences that need immediate feedback. Google is clearly trying to make fast frontier behavior the default engine for agentic workflows.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"\u002F\u002F Classic model usage\\nanswer = model.generate(\\\"Explain this function\\\") \u002F\u002F Agentic model usage\\nplan = model.plan(task)\\ntoolResult = tools.run(plan.nextStep)\\nverification = model.verify(toolResult) if verification.failed: rollback()\\nelse: continueExecution()\"},\"type\":\"code\"},{\"data\":{\"text\":\"That shift is why Gemini 3.5 belongs in the same cluster as Antigravity. Antigravity needs a model that can plan, execute, verify, and keep moving without turning the development environment into a waiting room. This is where compute performance becomes developer experience.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"Why Speed and Action Belong Together\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Google’s keynote framing around speed is not just marketing. Agentic systems multiply latency. A single answer has one model round trip. A multi-agent workflow can have many: planning, tool selection, code editing, terminal execution, verification, repair, summarization, and human review. If each step is slow, the whole system becomes unusable.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"Fast output improves interactive coding loops.\",\"Low latency makes multi-step agents feel less brittle.\",\"High throughput matters when agents run in background maintenance tasks.\",\"Stable reasoning matters because faster wrong actions are still wrong actions.\",\"Verification remains essential because speed alone does not equal trust.\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"The useful agent is not the model that answers once. It is the model that can keep acting without making the workflow feel like a queue.\",\"caption\":\"Agentic runtime perspective\"},\"type\":\"quote\"},{\"data\":{\"text\":\"Gemini Omni vs Gemini 3.5: Different Jobs, Same Stack\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The cleanest way to understand the I\u002FO 2026 compute announcements is to separate the jobs. Gemini Omni is a multimodal creation and editing model. Gemini 3.5 Flash is an action-oriented model for agents and coding. They are complementary, not interchangeable. A product team might use Omni to generate media and use Gemini 3.5 Flash to orchestrate the workflow around that media.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"\u002F\u002F Product workflow example\\ncreativeAsset = GeminiOmni.generateVideo(brief) agentPlan = Gemini35Flash.plan({ task: \\\"Publish launch campaign\\\", asset: creativeAsset, channels: [\\\"YouTube Shorts\\\", \\\"landing page\\\", \\\"email\\\"]\\n}) Gemini35Flash.execute(agentPlan, tools)\"},\"type\":\"code\"},{\"data\":{\"text\":\"That is the broader architecture Google is pointing toward. Creative models generate richer artifacts. Action models coordinate tasks around those artifacts. Product surfaces expose those capabilities to users. Developer tools provide orchestration and verification. Device surfaces make the assistant available in more contexts.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"What This Means for Developers\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Developers should not treat Gemini 3.5 Flash as only another model option. It changes the shape of applications that can be built on top of Google’s stack. If a model can handle longer task sequences, use tools reliably, and operate fast enough for real workflows, then applications can shift from isolated AI features to model-mediated processes.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"Code agents can refactor, test, and verify larger task units.\",\"Product agents can help users move from intent to action inside one flow.\",\"Media applications can combine Omni generation with Gemini 3.5 orchestration.\",\"Search and Workspace features can move from answering toward executing.\",\"Mobile and XR surfaces can use Gemini as a live context layer instead of a distant cloud assistant.\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"The danger is overbuilding before the operational model is mature. Developers still need versioning, evaluation, logging, cost controls, fallback behavior, and human confirmation for high-impact actions. A stronger model does not remove the need for engineering discipline. It increases the value of that discipline.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"What This Means for Product Teams\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Product teams should read Gemini Omni and Gemini 3.5 as a signal that Google wants AI to become an execution layer across the ecosystem. That creates opportunity and risk. The opportunity is a faster path from user intent to finished output. The risk is that product behavior becomes less transparent if agent decisions are hidden behind a polished interface.\"},\"type\":\"paragraph\"},{\"data\":{\"code\":\"agenticProductChecklist = { userIntent: \\\"clear\\\", modelRole: \\\"defined\\\", tools: \\\"permissioned\\\", outputs: \\\"reviewable\\\", failures: \\\"recoverable\\\", costs: \\\"measured\\\", userControl: \\\"explicit\\\"\\n}\"},\"type\":\"code\"},{\"data\":{\"text\":\"If a product team cannot answer what the model is allowed to do, how it is evaluated, and how the user can recover from a bad action, the integration is not ready for production. This is especially true when Gemini is connected to Search, Workspace, Shopping, Android, or wearable interfaces.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"The Reality Check: Benchmarks Are Not the Product\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"The risk with every model launch is benchmark intoxication. Better performance matters, but production value depends on integration quality. A model can be faster and stronger while still failing a specific workflow because the prompts are weak, tool policies are loose, evaluation is shallow, or rollback paths are missing.\"},\"type\":\"paragraph\"},{\"data\":{\"items\":[\"A creative model needs consistency, editability, and output control.\",\"An action model needs tool discipline, verification, and safe failure modes.\",\"A developer model needs repeatable behavior inside real repositories.\",\"A product model needs user trust and clear control boundaries.\",\"A platform model needs versioning, monitoring, and cost observability.\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"The launch model gets attention. The operating model decides whether it survives production.\",\"caption\":\"Engineering reality check\"},\"type\":\"quote\"},{\"data\":{\"text\":\"How This Fits the Google I\u002FO 2026 Cluster\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"This article covers the model layer of the cluster. The \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\\\">main Google I\u002FO 2026 hub\u003C\u002Fa> explains the full architectural shift. \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\\\">Antigravity, AI Studio, and Google DevTools\u003C\u002Fa> explains how the models enter developer workflows. \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\\\">Android XR and Intelligent Eyewear\u003C\u002Fa> explains how the same intelligence moves into wearable surfaces. \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\\\">Agentic Products Across Search, Workspace, and Shopping\u003C\u002Fa> explains what happens when these capabilities reach consumer and productivity products.\"},\"type\":\"paragraph\"},{\"data\":{\"text\":\"Related Articles in This Cluster\",\"level\":2},\"type\":\"header\"},{\"data\":{\"items\":[\"\u003Cb>Main Hub:\u003C\u002Fb> \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\\\">Google I\u002FO 2026: Architectural Pivots, Agentic AI, and the Unified Ecosystem Reality Check\u003C\u002Fa>\",\"\u003Cb>Developer Tooling:\u003C\u002Fb> \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\\\">Google I\u002FO 2026: Antigravity, AI Studio, and Google DevTools\u003C\u002Fa>\",\"\u003Cb>Android, XR, and Device Surfaces:\u003C\u002Fb> \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\\\">Google I\u002FO 2026: Android XR and Intelligent Eyewear\u003C\u002Fa>\",\"\u003Cb>Agentic Consumer Products:\u003C\u002Fb> \u003Ca href=\\\"\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\\\">Google I\u002FO 2026: Agentic Products Across Search, Workspace, and Shopping\u003C\u002Fa>\"],\"style\":\"unordered\"},\"type\":\"list\"},{\"data\":{\"text\":\"Final Perspective\",\"level\":2},\"type\":\"header\"},{\"data\":{\"text\":\"Gemini Omni and Gemini 3.5 define the compute split behind Google I\u002FO 2026. Omni is the creation layer: multimodal input, video output, conversational editing, and world-grounded generation. Gemini 3.5 Flash is the action layer: fast frontier intelligence for agents, coding, and long-horizon tasks. Together, they explain why the rest of the keynote looks the way it does. Antigravity needs action-grade models. Search and Workspace need agents that can reason and execute. Android XR needs live context understanding. The keynote is not just a model story, but without this compute layer the rest of the story does not work.\"},\"type\":\"paragraph\"},{\"data\":{},\"type\":\"delimiter\"}],\"version\":\"2.30.8\"}",{"time":423,"blocks":424,"version":581},1774828800000,[425,428,431,434,437,440,447,451,454,457,460,463,466,469,472,475,478,481,484,487,495,499,502,505,508,511,514,517,525,528,531,534,536,539,542,545,553,557,560,563,566,573,576,579],{"data":426,"type":233},{"text":427},"\u003Cb>Google I\u002FO 2026\u003C\u002Fb> put the compute layer of Google’s AI strategy back in the center. The keynote was full of agents, developer tools, wearable interfaces, and product integrations, but all of those layers depend on two model directions: \u003Cb>Gemini Omni\u003C\u002Fb> and \u003Cb>Gemini 3.5\u003C\u002Fb>. One pushes multimodal creation and editing. The other pushes frontier intelligence with action, especially through Gemini 3.5 Flash.",{"data":429,"type":233},{"text":430},"This article is the \u003Cb>compute-layer spoke\u003C\u002Fb> inside the larger \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026 architectural hub\u003C\u002Fa>. The developer-tooling side is covered in \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Antigravity, AI Studio, and Google DevTools\u003C\u002Fa>. The device-surface story is covered in \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Android XR and Intelligent Eyewear\u003C\u002Fa>. The product-surface story is covered in \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">Agentic Products Across Search, Workspace, and Shopping\u003C\u002Fa>.",{"data":432,"type":233},{"text":433},"The important thing is to avoid reading Gemini Omni and Gemini 3.5 as one generic “new AI model” announcement. They solve different product and infrastructure problems. Omni is about any-input creative generation, starting with video. Gemini 3.5 is about fast, action-oriented intelligence for agents, coding, and long-horizon workflows. That separation matters for developers, product teams, and platform architects.",{"data":435,"type":42},{"text":436,"level":243},"The Compute Layer Behind the Keynote",{"data":438,"type":233},{"text":439},"Google’s own I\u002FO 2026 collection frames the event around two new models: Gemini Omni and Gemini 3.5. Gemini Omni is described as a leap forward in world understanding, multimodality, and editing, able to create output from input across text, images, audio, and video, starting with video. Gemini 3.5 Flash is introduced as the first model in the Gemini 3.5 family, combining frontier intelligence with action.",{"data":441,"type":255},{"items":442,"style":254},[443,444,445,446],"Gemini Omni focuses on multimodal creation and editing, starting with video.","Gemini 3.5 Flash focuses on action-grade speed for agents, coding, and complex workflows.","The model split mirrors Google’s broader platform split: create richer media, then act faster across products and tools.","The real strategic value comes when these models are connected to Antigravity, AI Studio, Search, Workspace, Android XR, and Gemini-powered devices.",{"data":448,"type":260},{"text":449,"caption":450},"The model story is not just bigger context or better demos. It is role specialization: creation on one side, action on the other.","Compute-layer reading",{"data":452,"type":42},{"text":453,"level":243},"Gemini Omni: Creation From Any Input",{"data":455,"type":233},{"text":456},"Gemini Omni is Google’s most direct attempt to fuse reasoning with creation. The official positioning is clear: it can create anything from any input, starting with video. The model accepts combinations of text, image, audio, and video, then generates high-quality video grounded in Gemini’s real-world knowledge. It also supports conversational video editing, which is likely to matter more than one-shot generation in actual creative workflows.",{"data":458,"type":233},{"text":459},"That last point is important. The practical value of creative AI is rarely only “generate a video.” Production workflows require revision, replacement, timing, style continuity, and conversation around changes. If Omni can make editing more conversational while preserving context, it becomes more than a novelty generator. It becomes a creative interface.",{"data":461,"type":273},{"code":462},"\u002F\u002F Gemini Omni should be read as a creative workflow model\ninput = { text: \"Show a product launch in a rainy cyberpunk city\", image: \"brand-reference.png\", audio: \"voiceover.wav\", video: \"rough-cut.mp4\"\n} output = GeminiOmni.generateVideo(input)\nrevision = GeminiOmni.edit(output, \"Make the lighting warmer and keep the same product design\")",{"data":464,"type":233},{"text":465},"For teams building media tools, ecommerce product visuals, education content, game prototyping, advertising workflows, or social video pipelines, this is the part of I\u002FO 2026 that matters most. The interface moves from manual timeline editing toward intent-based multimodal editing. That does not remove creative judgment, but it changes where creative labor is spent.",{"data":467,"type":42},{"text":468,"level":243},"Gemini 3.5 Flash: Frontier Intelligence With Action",{"data":470,"type":233},{"text":471},"Gemini 3.5 Flash is the other half of the compute story. Google describes the Gemini 3.5 family as combining frontier intelligence with action, and positions 3.5 Flash as the first release in that family. The emphasis is on agents, coding, complex long-horizon tasks, and real-world utility. That language is not accidental. It points directly toward systems that do things, not just systems that answer.",{"data":473,"type":233},{"text":474},"For developers, “Flash” matters because agentic workflows are latency-sensitive. A model that is smart but slow can work for research or careful writing. It is much harder to fit into terminal-native automation, IDE loops, multi-step tool usage, background agents, and product experiences that need immediate feedback. Google is clearly trying to make fast frontier behavior the default engine for agentic workflows.",{"data":476,"type":273},{"code":477},"\u002F\u002F Classic model usage\nanswer = model.generate(\"Explain this function\") \u002F\u002F Agentic model usage\nplan = model.plan(task)\ntoolResult = tools.run(plan.nextStep)\nverification = model.verify(toolResult) if verification.failed: rollback()\nelse: continueExecution()",{"data":479,"type":233},{"text":480},"That shift is why Gemini 3.5 belongs in the same cluster as Antigravity. Antigravity needs a model that can plan, execute, verify, and keep moving without turning the development environment into a waiting room. This is where compute performance becomes developer experience.",{"data":482,"type":42},{"text":483,"level":243},"Why Speed and Action Belong Together",{"data":485,"type":233},{"text":486},"Google’s keynote framing around speed is not just marketing. Agentic systems multiply latency. A single answer has one model round trip. A multi-agent workflow can have many: planning, tool selection, code editing, terminal execution, verification, repair, summarization, and human review. If each step is slow, the whole system becomes unusable.",{"data":488,"type":255},{"items":489,"style":254},[490,491,492,493,494],"Fast output improves interactive coding loops.","Low latency makes multi-step agents feel less brittle.","High throughput matters when agents run in background maintenance tasks.","Stable reasoning matters because faster wrong actions are still wrong actions.","Verification remains essential because speed alone does not equal trust.",{"data":496,"type":260},{"text":497,"caption":498},"The useful agent is not the model that answers once. It is the model that can keep acting without making the workflow feel like a queue.","Agentic runtime perspective",{"data":500,"type":42},{"text":501,"level":243},"Gemini Omni vs Gemini 3.5: Different Jobs, Same Stack",{"data":503,"type":233},{"text":504},"The cleanest way to understand the I\u002FO 2026 compute announcements is to separate the jobs. Gemini Omni is a multimodal creation and editing model. Gemini 3.5 Flash is an action-oriented model for agents and coding. They are complementary, not interchangeable. A product team might use Omni to generate media and use Gemini 3.5 Flash to orchestrate the workflow around that media.",{"data":506,"type":273},{"code":507},"\u002F\u002F Product workflow example\ncreativeAsset = GeminiOmni.generateVideo(brief) agentPlan = Gemini35Flash.plan({ task: \"Publish launch campaign\", asset: creativeAsset, channels: [\"YouTube Shorts\", \"landing page\", \"email\"]\n}) Gemini35Flash.execute(agentPlan, tools)",{"data":509,"type":233},{"text":510},"That is the broader architecture Google is pointing toward. Creative models generate richer artifacts. Action models coordinate tasks around those artifacts. Product surfaces expose those capabilities to users. Developer tools provide orchestration and verification. Device surfaces make the assistant available in more contexts.",{"data":512,"type":42},{"text":513,"level":243},"What This Means for Developers",{"data":515,"type":233},{"text":516},"Developers should not treat Gemini 3.5 Flash as only another model option. It changes the shape of applications that can be built on top of Google’s stack. If a model can handle longer task sequences, use tools reliably, and operate fast enough for real workflows, then applications can shift from isolated AI features to model-mediated processes.",{"data":518,"type":255},{"items":519,"style":254},[520,521,522,523,524],"Code agents can refactor, test, and verify larger task units.","Product agents can help users move from intent to action inside one flow.","Media applications can combine Omni generation with Gemini 3.5 orchestration.","Search and Workspace features can move from answering toward executing.","Mobile and XR surfaces can use Gemini as a live context layer instead of a distant cloud assistant.",{"data":526,"type":233},{"text":527},"The danger is overbuilding before the operational model is mature. Developers still need versioning, evaluation, logging, cost controls, fallback behavior, and human confirmation for high-impact actions. A stronger model does not remove the need for engineering discipline. It increases the value of that discipline.",{"data":529,"type":42},{"text":530,"level":243},"What This Means for Product Teams",{"data":532,"type":233},{"text":533},"Product teams should read Gemini Omni and Gemini 3.5 as a signal that Google wants AI to become an execution layer across the ecosystem. That creates opportunity and risk. The opportunity is a faster path from user intent to finished output. The risk is that product behavior becomes less transparent if agent decisions are hidden behind a polished interface.",{"data":535,"type":273},{"code":347},{"data":537,"type":233},{"text":538},"If a product team cannot answer what the model is allowed to do, how it is evaluated, and how the user can recover from a bad action, the integration is not ready for production. This is especially true when Gemini is connected to Search, Workspace, Shopping, Android, or wearable interfaces.",{"data":540,"type":42},{"text":541,"level":243},"The Reality Check: Benchmarks Are Not the Product",{"data":543,"type":233},{"text":544},"The risk with every model launch is benchmark intoxication. Better performance matters, but production value depends on integration quality. A model can be faster and stronger while still failing a specific workflow because the prompts are weak, tool policies are loose, evaluation is shallow, or rollback paths are missing.",{"data":546,"type":255},{"items":547,"style":254},[548,549,550,551,552],"A creative model needs consistency, editability, and output control.","An action model needs tool discipline, verification, and safe failure modes.","A developer model needs repeatable behavior inside real repositories.","A product model needs user trust and clear control boundaries.","A platform model needs versioning, monitoring, and cost observability.",{"data":554,"type":260},{"text":555,"caption":556},"The launch model gets attention. The operating model decides whether it survives production.","Engineering reality check",{"data":558,"type":42},{"text":559,"level":243},"How This Fits the Google I\u002FO 2026 Cluster",{"data":561,"type":233},{"text":562},"This article covers the model layer of the cluster. The \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">main Google I\u002FO 2026 hub\u003C\u002Fa> explains the full architectural shift. \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Antigravity, AI Studio, and Google DevTools\u003C\u002Fa> explains how the models enter developer workflows. \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Android XR and Intelligent Eyewear\u003C\u002Fa> explains how the same intelligence moves into wearable surfaces. \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">Agentic Products Across Search, Workspace, and Shopping\u003C\u002Fa> explains what happens when these capabilities reach consumer and productivity products.",{"data":564,"type":42},{"text":565,"level":243},"Related Articles in This Cluster",{"data":567,"type":255},{"items":568,"style":254},[569,570,571,572],"\u003Cb>Main Hub:\u003C\u002Fb> \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-architectural-pivots-agentic-ai-and-the-unified-ecosystem-reality-check\">Google I\u002FO 2026: Architectural Pivots, Agentic AI, and the Unified Ecosystem Reality Check\u003C\u002Fa>","\u003Cb>Developer Tooling:\u003C\u002Fb> \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-antigravity-ai-studio-and-google-devtools\">Google I\u002FO 2026: Antigravity, AI Studio, and Google DevTools\u003C\u002Fa>","\u003Cb>Android, XR, and Device Surfaces:\u003C\u002Fb> \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear\">Google I\u002FO 2026: Android XR and Intelligent Eyewear\u003C\u002Fa>","\u003Cb>Agentic Consumer Products:\u003C\u002Fb> \u003Ca href=\"\u002Fblog\u002Fgoogle-io-2026-agentic-products-search-workspace-and-shopping\">Google I\u002FO 2026: Agentic Products Across Search, Workspace, and Shopping\u003C\u002Fa>",{"data":574,"type":42},{"text":575,"level":243},"Final Perspective",{"data":577,"type":233},{"text":578},"Gemini Omni and Gemini 3.5 define the compute split behind Google I\u002FO 2026. Omni is the creation layer: multimodal input, video output, conversational editing, and world-grounded generation. Gemini 3.5 Flash is the action layer: fast frontier intelligence for agents, coding, and long-horizon tasks. Together, they explain why the rest of the keynote looks the way it does. Antigravity needs action-grade models. Search and Workspace need agents that can reason and execute. Android XR needs live context understanding. The keynote is not just a model story, but without this compute layer the rest of the story does not work.",{"data":580,"type":393},{},"2.30.8","Google I\u002FO 2026 put Gemini Omni and Gemini 3.5 at the center of Google’s agentic AI strategy. This article breaks down the difference between multimodal creation and action-grade intelligence, why Gemini 3.5 Flash matters for agents and coding, and how these models power the wider Google I\u002FO 2026 platform shift.",{"lang":7,"title":224,"content":226,"contentJson":584,"excerpt":395},{"time":228,"blocks":585,"version":394},[586,588,590,592,594,596,599,601,603,605,607,609,611,613,615,617,619,621,623,625,628,630,632,634,636,638,640,642,645,647,649,651,653,655,657,659,662,664,666,668,670,673,675,677],{"data":587,"type":233},{"text":232},{"data":589,"type":233},{"text":236},{"data":591,"type":233},{"text":239},{"data":593,"type":42},{"text":242,"level":243},{"data":595,"type":233},{"text":246},{"data":597,"type":255},{"items":598,"style":254},[250,251,252,253],{"data":600,"type":260},{"text":258,"caption":259},{"data":602,"type":42},{"text":263,"level":243},{"data":604,"type":233},{"text":266},{"data":606,"type":233},{"text":269},{"data":608,"type":273},{"code":272},{"data":610,"type":233},{"text":276},{"data":612,"type":42},{"text":279,"level":243},{"data":614,"type":233},{"text":282},{"data":616,"type":233},{"text":285},{"data":618,"type":273},{"code":288},{"data":620,"type":233},{"text":291},{"data":622,"type":42},{"text":294,"level":243},{"data":624,"type":233},{"text":297},{"data":626,"type":255},{"items":627,"style":254},[301,302,303,304,305],{"data":629,"type":260},{"text":308,"caption":309},{"data":631,"type":42},{"text":312,"level":243},{"data":633,"type":233},{"text":315},{"data":635,"type":273},{"code":318},{"data":637,"type":233},{"text":321},{"data":639,"type":42},{"text":324,"level":243},{"data":641,"type":233},{"text":327},{"data":643,"type":255},{"items":644,"style":254},[331,332,333,334,335],{"data":646,"type":233},{"text":338},{"data":648,"type":42},{"text":341,"level":243},{"data":650,"type":233},{"text":344},{"data":652,"type":273},{"code":347},{"data":654,"type":233},{"text":350},{"data":656,"type":42},{"text":353,"level":243},{"data":658,"type":233},{"text":356},{"data":660,"type":255},{"items":661,"style":254},[360,361,362,363,364],{"data":663,"type":260},{"text":367,"caption":368},{"data":665,"type":42},{"text":371,"level":243},{"data":667,"type":233},{"text":374},{"data":669,"type":42},{"text":377,"level":243},{"data":671,"type":255},{"items":672,"style":254},[381,382,383,384],{"data":674,"type":42},{"text":387,"level":243},{"data":676,"type":233},{"text":390},{"data":678,"type":393},{},"Post erfolgreich abgerufen",{"items":681,"source":766,"manualIds":767,"manualMatchedIds":768},[682,689,696,703,710,717,724,731,738,745,752,759],{"id":683,"slug":684,"title":685,"excerpt":686,"featuredImage":687,"publishedAt":688},"448","google-io-2026-android-xr-and-intelligent-eyewear","Google I\u002FO 2026：Android XR、智能眼镜与环境AI界面","Google I\u002FO 2026 将 Android XR 和智能眼镜从概念推向实际平台方向。本文解析了音频眼镜、显示眼镜、Gemini 驱动的上下文感知、开发者影响、隐私风险，以及为何可穿戴 AI 更关乎创造环境辅助界面，而非取代手机。","\u002Fuploads\u002F2026\u002F05\u002Fgoogle-io-2026-android-xr-and-intelligent-eyewear-1779227942270-dtsm9y.webp","2026-05-21T11:05:00.000Z",{"id":690,"slug":691,"title":692,"excerpt":693,"featuredImage":694,"publishedAt":695},"434","evaluation-harness","全面评估指南：精通LLM性能评估","本指南详细介绍了评估工具（Evaluation Harness），这是一个在企业级LLMOps流程中严格评估大型语言模型（LLM）能力的关键框架。您将学习其设置方法、最佳实践以及高级技巧，以确保模型基准测试与优化的可靠性。","\u002Fuploads\u002F2026\u002F04\u002Fevaluation-harness-1775466944495-4s0xv2.webp","2026-03-01T17:50:00.000Z",{"id":697,"slug":698,"title":699,"excerpt":700,"featuredImage":701,"publishedAt":702},"452","zbt-z8102ax-5g-openwrt-router-review-dual-sim-rm500u-ea-and-an-honest-assessment","ZBT Z8102AX 5G OpenWrt路由器评测：双SIM卡、RM500U-EA及真实评估","ZBT Z8102AX是一款独特的5G路由器，采用OpenWrt基础系统、双SIM卡设计以及Quectel RM500U-EA调制解调器。在测试中，它在灵活性、接口和移动连接方面展现出明显优势，但也暴露出厂商定制版OpenWrt固件的典型缺陷。","\u002Fuploads\u002F2026\u002F06\u002Fopenwrt-router-review-dual-sim-01-1781620588908-fwmzj7.webp","2026-06-16T07:34:00.000Z",{"id":704,"slug":705,"title":706,"excerpt":707,"featuredImage":708,"publishedAt":709},"492","mcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits","MCP 解析：它连接什么、不做什么以及它适用于何处","模型上下文协议通过标准的客户端-服务器边界，将AI应用程序连接到外部工具、资源和提示。了解MCP能做什么、不能做什么，以及它在智能体架构中的定位。","\u002Fuploads\u002F2026\u002F10\u002Fmcp-explained-what-it-connects-what-it-does-not-do-and-where-it-fits-1791486640275-7ub1cq.webp","2026-10-08T15:09:00.000Z",{"id":711,"slug":712,"title":713,"excerpt":714,"featuredImage":715,"publishedAt":716},"487","vector-databases-embeddings-and-reranking-three-different-parts-of-retrieval","向量数据库、嵌入和重排序：检索的三个不同部分","嵌入表示含义，向量数据库检索候选结果，重排序器则精炼结果。了解这三个检索层在RAG中如何不同并协同工作。","\u002Fuploads\u002F2026\u002F10\u002Fvector-databases-embeddings-and-reranking-three-different-parts-of-retrieval-1791480129884-9dtasz.webp","2026-10-08T11:21:00.000Z",{"id":718,"slug":719,"title":720,"excerpt":721,"featuredImage":722,"publishedAt":723},"453","zbt-z8102ax-openwrt-2102-firmware-review","ZBT Z8102AX OpenWrt 21.02 固件评测：足够稳定，但能否面向未来？","ZBT Z8102AX 运行的是厂商修改版 OpenWrt 21.02 固件，内核版本为 5.4.246。实际测试中，该固件运行稳定，可保持路由器连续数日正常工作，但老旧的基础版本引发了关于安全性、调制解调器控制、升级路径及长期可维护性的重要问题。","\u002Fuploads\u002F2026\u002F06\u002Fopenwrt-router-review-dual-sim-04-1781620594653-uvf5oy.webp","2026-06-16T10:39:00.000Z",{"id":725,"slug":726,"title":727,"excerpt":728,"featuredImage":729,"publishedAt":730},"454","zbt-z8102ax-rm500u-ea-5g-modem-test","Quectel RM500U-EA在ZBT Z8102AX中：5G频段、o2德国及实际信号表现","ZBT Z8102AX 使用移远 RM500U-EA 调制解调器实现 4G 和 5G 连接。在首次实际测试中，该路由器成功连接至德国 o2 网络，使用 LTE Band 3 和 NR n28 频段。调制解调器工作正常，但更深层次的诊断功能如 RSRP、RSRQ、SINR、频段锁定及小区行为仍需进一步测试。","\u002Fuploads\u002F2026\u002F06\u002Fopenwrt-router-review-dual-sim-06-1781620597879-qay2sx.webp","2026-06-16T08:39:00.000Z",{"id":732,"slug":733,"title":734,"excerpt":735,"featuredImage":736,"publishedAt":737},"489","agentic-ai-explained-when-an-ai-system-can-plan-use-tools-and-act","智能体AI解析：当AI系统能够规划、使用工具并采取行动","代理式AI在多步执行循环中使用模型，这些模型可以在明确的运行时和权限边界内选择工具、观察结果、更新状态并调整其下一步行动。","\u002Fuploads\u002F2026\u002F10\u002Fagentic-ai-explained-when-an-ai-system-can-plan-use-tools-and-act-1791481499084-wnji2a.webp","2026-10-08T11:43:00.000Z",{"id":739,"slug":740,"title":741,"excerpt":742,"featuredImage":743,"publishedAt":744},"460","ai-agent-reliability-why-the-final-answer-is-not-enough","AI Agent可靠性：为什么最终答案并不足够","正确的输出并不能证明推理的正确性、执行的安全性，或系统的可信赖性。","\u002Fuploads\u002F2026\u002F09\u002Fai-agent-reliability-why-the-final-answer-is-not-enough-1788955466306-pl0qhz.webp","2026-09-09T04:01:00.000Z",{"id":746,"slug":747,"title":748,"excerpt":749,"featuredImage":750,"publishedAt":751},"386","enterprise-start-here-your-gateway-to-operational-excellence","Enterprise Start Here: Your Gateway to Operational Excellence","New to our enterprise platform? This guide provides a structured onboarding path, from foundational reference models to actionable playbooks, runbooks, and assessments designed for seamless implementation.","\u002Fuploads\u002F2026\u002F02\u002Ffrom-global-business-to-the-kitchen-a-reverse-communication-system-that-still-scales-1771251820950-zhp2bz.webp","2026-03-01T11:55:00.000Z",{"id":753,"slug":754,"title":755,"excerpt":756,"featuredImage":757,"publishedAt":758},"480","when-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger","人工智能何时应停止信任自身知识？——检索触发机制","AI 模型并非每个问题都需要检索。重要的问题在于知道何时其内部知识已不再足够。检索触发器是一个实用的决策边界，它决定 AI 系统何时应停止仅依赖模型知识，并在回答前获取外部证据。","\u002Fuploads\u002F2026\u002F09\u002Fwhen-should-an-ai-stop-trusting-its-own-knowledge-the-retrieval-trigger-1790574991244-f4rpyg.webp","2026-09-28T01:49:00.000Z",{"id":760,"slug":761,"title":762,"excerpt":763,"featuredImage":764,"publishedAt":765},"467","the-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers","答案有效性边界：相关性到可靠AI答案之间缺失的层级","一个来源可能相关、权威，但对于所提出的问题仍然是错误的。缺失的层次是适用性：答案成立的条件，以及迫使其被重新考虑的变化。本文介绍了“答案有效性边界”这一面向人类、AI搜索和RAG系统的来源设计模式。","\u002Fuploads\u002F2026\u002F09\u002Fthe-answer-validity-boundary-the-missing-layer-between-relevance-and-reliable-ai-answers-1790272901306-1g5jly.webp","2026-09-24T11:59:00.000Z","fallback",[],[]]