[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-microsoft-mai-models-build-2026-zh":3,"article-related-microsoft-mai-models-build-2026-zh":33,"series-model-release-0bb91791-f4b6-4d51-899c-6eeb239f942a":82},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":25,"views":29,"created_at":30,"published_at":31,"topic_cluster_id":32},"0bb91791-f4b6-4d51-899c-6eeb239f942a","microsoft-mai-models-build-2026-zh","Microsoft把 Copilot 拉回主場","\u003Cp data-speakable=\"summary\">\u003Ca href=\"\u002Ftag\u002Fmicrosoft\">Microsoft\u003C\u002Fa> 在 Build \u003Ca href=\"\u002Fnews\u002Fmicrosoft-build-2026-agents-into-systems-zh\">2026\u003C\u002Fa> 推出 7 款 MAI 模型，主打 MAI-Thinking-1，還把 Scout、Surface RTX Spark 和 Windows 容器一起端上桌。\u003C\u002Fp>\u003Cp>說真的，這場發表會很直白。Microsoft 不想再只當 \u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa> 的轉接站。它要把模型、產品和硬體都收回自己手上。\u003C\u002Fp>\u003Cp>這次亮點很集中。MAI-Thinking-1 有 350 億 active parameters，context window 也拉到 128K。對開發者來說，這代表長提示、程式碼和多步推理，會更像一個完整工作流。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>數字\u003C\u002Fth>\u003Cth>狀態\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>MAI-Thinking-1\u003C\u002Ftd>\u003Ctd>35B active parameters，128K context\u003C\u002Ftd>\u003Ctd>Microsoft Foundry private preview\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>MAI-Transcribe-1.5\u003C\u002Ftd>\u003Ctd>43 種語言\u003C\u002Ftd>\u003Ctd>即將推出\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>MAI-Voice-2\u003C\u002Ftd>\u003Ctd>新增 15 種語言\u003C\u002Ftd>\u003Ctd>已可用\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Surface RTX Spark Dev Box\u003C\u002Ftd>\u003Ctd>最高 1 petaflop，128GB unified memory\u003C\u002Ftd>\u003Ctd>今年稍晚美國開賣\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>本機模型支援\u003C\u002Ftd>\u003Ctd>最高 120B 參數\u003C\u002Ftd>\u003Ctd>規劃中\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Microsoft 這次是自己做整套\u003C\u002Fh2>\u003Cp>過去兩年，很多人提到 \u003Ca href=\"\u002Ftag\u002Fcopilot\">Copilot\u003C\u002Fa>，就會先想到 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002F\" target=\"_blank\" rel=\"noopener\">OpenAI\u003C\u002Fa>。\u003Ca href=\"https:\u002F\u002Fbuild.microsoft.com\u002F\" target=\"_blank\" rel=\"noopener\">Build 2026\u003C\u002Fa> 直接把這層關係往後推。Microsoft 這次講得很明白：它要自己做文本、語音、轉錄、圖片和程式碼模型。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781330585064-s9ya.png\" alt=\"Microsoft把 Copilot 拉回主場\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這件事很現實。模型握在自己手上，價格、延遲、整合方式都比較好控。對企業客戶來說，這也比較像一條完整供應鏈，而不是到處拼 API。\u003C\u002Fp>\u003Cp>Microsoft 的命名也很務實。MAI-Thinking-1、MAI-Image-2.5、MAI-Transcribe-1.5、MAI-Voice-2、MAI-\u003Ca href=\"\u002Fnews\u002Ffable-5-claude-code-like-coworker-zh\">Code\u003C\u002Fa>-1。名字不花俏，但至少一看就知道用途。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cstrong>MAI-Thinking-1\u003C\u002Fstrong>：推理與 coding\u003C\u002Fli>\u003Cli>\u003Cstrong>MAI-Image-2.5\u003C\u002Fstrong>：圖片生成\u003C\u002Fli>\u003Cli>\u003Cstrong>MAI-Transcribe-1.5\u003C\u002Fstrong>：語音轉文字\u003C\u002Fli>\u003Cli>\u003Cstrong>MAI-Voice-2\u003C\u002Fstrong>：文字轉語音\u003C\u002Fli>\u003Cli>\u003Cstrong>MAI-Code-1\u003C\u002Fstrong>：程式輔助\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這種切法其實很聰明。因為大多數團隊，不需要一顆萬能模型。大家要的是可預測、可控、可塞進產品裡的工具。\u003C\u002Fp>\u003Ch2>MAI-Thinking-1 才是主菜\u003C\u002Fh2>\u003Cp>七個模型裡，最值得盯的就是 MAI-Thinking-1。Microsoft Developer CMO 兼 GitHub COO Kyle Daigle 說，這顆模型是為了複雜多步驟指令、長上下文推理和 code generation 而做。\u003C\u002Fp>\u003Cp>Microsoft 還說，外部評測者比較偏好 MAI-Thinking-1，對手是 \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-sonnet-4-5\" target=\"_blank\" rel=\"noopener\">Anthropic 的 Claude Sonnet 4.6\u003C\u002Fa>。它也在 SWE Bench Pro coding \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> 上追平 Claude Opus 4.6。這種說法很硬，因為它不是只拿來做 demo。\u003C\u002Fp>\u003Cp>更有意思的是資料來源。Microsoft 說這顆模型是用商業授權資料訓練的。講白了，這是在避開很多模型廠商都碰過的授權爭議。\u003C\u002Fp>\u003Cblockquote>“MAI-Thinking-1 was designed to be good at complex multi-step instructions, long context reasoning, and code generation,” said Kyle Daigle, Microsoft Developer CMO and COO of GitHub.\u003C\u002Fblockquote>\u003Cp>當然，開發者還是別太快高潮。Microsoft 目前沒把完整 model card 全部攤開。單看一條 benchmark，不代表所有場景都贏。\u003C\u002Fp>\u003Cp>但 128K context、Foundry private preview、還有 Microsoft 內部產品線一起導入，這組合已經夠有看頭了。這不是試水溫而已。\u003C\u002Fp>\u003Ch2>其他六個模型是在補產品空缺\u003C\u002Fh2>\u003Cp>如果只看 MAI-Thinking-1，你會以為 Microsoft 只想做一顆大模型。其實不是。它把語音、轉錄、圖片、coding 都補齊了。這比較像\u003Ca href=\"\u002Fnews\u002Fchatgpt-grew-from-chatbot-to-platform-zh\">平台\u003C\u002Fa>策略。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781330578715-8baj.png\" alt=\"Microsoft把 Copilot 拉回主場\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>企業導入 AI 時，最常卡住的不是最強模型，而是周邊功能。轉錄準不準、語音自然不自然、圖片能不能直接塞進簡報，這些才是每天會碰到的事。\u003C\u002Fp>\u003Cp>而且 Microsoft 的部署節奏很清楚。部分模型已經進了日常產品，部分則放進 Foundry 和 MAI Playground。這種做法比較像把模型當零件，而不是只拿來展示。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fmicrosoft-365\u002Fpowerpoint\" target=\"_blank\" rel=\"noopener\">PowerPoint\u003C\u002Fa> 和 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fmicrosoft-365\u002Fonedrive\u002Fonline-cloud-storage\" target=\"_blank\" rel=\"noopener\">OneDrive\u003C\u002Fa> 已使用 MAI-Image-2.5\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fcode.visualstudio.com\u002F\" target=\"_blank\" rel=\"noopener\">VS Code\u003C\u002Fa> 和 Copilot 已有 MAI-Code-1\u003C\u002Fli>\u003Cli>MAI-Transcribe-1.5 支援 43 種語言\u003C\u002Fli>\u003Cli>MAI-Voice-2 先支援新增 15 種語言\u003C\u002Fli>\u003Cli>MAI-Thinking-1 已在 Foundry private preview\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這種分層很實際。產品端先吃得到，開發者端再慢慢補。對 Microsoft 來說，這樣比較容易把 AI 變成預設功能，而不是額外賣點。\u003C\u002Fp>\u003Cp>我覺得這比那種「一顆模型打天下」的敘事更像真的產品團隊。雖然沒那麼炫，但比較能落地。\u003C\u002Fp>\u003Ch2>Scout、Surface RTX Spark 和 Windows 都在同一條線上\u003C\u002Fh2>\u003Cp>模型只是第一層。Microsoft 這次還推了 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fmicrosoft-365\u002Fcopilot\u002Fmicrosoft-scout\" target=\"_blank\" rel=\"noopener\">Microsoft Scout\u003C\u002Fa>，這是一個偏主動式的個人 agent，會處理排程、會議準備和日常工作。它先給 Frontier customers 用，代表 Microsoft 想先看真實使用情境。\u003C\u002Fp>\u003Cp>硬體也沒缺席。\u003Ca href=\"https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fdata-center\u002Fproducts\u002Frtx-spark\u002F\" target=\"_blank\" rel=\"noopener\">Surface RTX Spark Dev Box\u003C\u002Fa> 用 NVIDIA RTX Spark，最高 1 petaflop，還有 128GB unified memory。Microsoft 說它能在本機跑到 120B 參數模型，這對在意延遲和資料外流的團隊很有吸引力。\u003C\u002Fp>\u003Cp>再往下看，Windows 也在改。\u003Ca href=\"https:\u002F\u002Flearn.microsoft.com\u002Fen-us\u002Fwindows\u002Fwin32\u002Fprocthread\u002Fcreating-processes-with-job-objects\" target=\"_blank\" rel=\"noopener\">Microsoft Execution Containers\u003C\u002Fa> 進入 preview。意思很簡單：agent 和本機模型，不只是跑程式而已，而是要像受控工作單元一樣運作。\u003C\u002Fp>\u003Cp>再加上 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fproject\u002Fmicrosoft-discovery\u002F\" target=\"_blank\" rel=\"noopener\">Microsoft Discovery\u003C\u002Fa> 已經 GA，整個方向就更清楚了。Microsoft 想管的是從雲端、桌面到本機的整條 AI 鏈。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cstrong>Scout\u003C\u002Fstrong>：處理 Microsoft 365 工作流程\u003C\u002Fli>\u003Cli>\u003Cstrong>Surface RTX Spark Dev Box\u003C\u002Fstrong>：本機推論與開發\u003C\u002Fli>\u003Cli>\u003Cstrong>Windows Execution Containers\u003C\u002Fstrong>：隔離 agent 行為\u003C\u002Fli>\u003Cli>\u003Cstrong>MAI models\u003C\u002Fstrong>：核心智慧層自己做\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這套組合很像在蓋自家城堡。不是只租別人的地，而是連地基都想自己掌握。\u003C\u002Fp>\u003Ch2>開發者真正該盯的是採用速度\u003C\u002Fh2>\u003Cp>現在的問題，不是 Microsoft 能不能再做出模型。問題是 MAI 會不會變成預設選項。只要 MAI-Code-1 在 \u003Ca href=\"\u002Ftag\u002Fvs-code\">VS Code\u003C\u002Fa> 持續出現，MAI-Image-2.5 持續進 PowerPoint 和 OneDrive，MAI-Thinking-1 也在 Foundry 穩定開放，Microsoft 就能慢慢降低對外部模型的依賴。\u003C\u002Fp>\u003Cp>對開發者來說，這有好有壞。好處是整合更單純。壞處是你會更綁 Microsoft 生態。這種便利和綁定，本來就是同一枚硬幣的兩面。\u003C\u002Fp>\u003Cp>我會先看兩件事。第一，private preview 會不會很快擴大。第二，MAI Playground 會不會真的好用，而不是只有發表會好看。如果這兩點成立，MAI 很可能變成 Microsoft 開發者的第一站。\u003C\u002Fp>\u003Cp>我的判斷很直接：Build 2026 最重要的，不是又多一顆模型，而是 Microsoft 開始把自家 AI 當基礎設施在經營。接下來就看它能不能把這套東西做得比外部方案更快、更省、更順手。","Microsoft 在 Build 2026 推出 7 款 MAI 模型，主打 MAI-Thinking-1、Scout、Surface RTX Spark 與 Windows 容器，想把 Copilot 和 AI 工具鏈拉回自家平台。","mashable.com","https:\u002F\u002Fmashable.com\u002Ftech\u002Fmicrosoft-launches-new-mai-family-of-models-at-build",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781330585064-s9ya.png","model-release","zh","3290c744-c64f-419f-8bfd-c87b597ad0d4",[17,18,19,20,21,22,23,24],"Microsoft","MAI models","Copilot","Build 2026","LLM","Foundry","Windows","Surface RTX Spark",[26,27,28],"Microsoft 在 Build 2026 推出 7 款 MAI 模型，重點是把模型掌握權拉回自家。","MAI-Thinking-1 有 35B active parameters 與 128K context，主打推理與 coding。","Scout、Surface RTX Spark 和 Windows 容器一起出現，代表 Microsoft 在做完整 AI 堆疊。",0,"2026-06-13T06:02:35.160901+00:00","2026-06-13T06:02:35.133+00:00","0ccb5d2e-69f1-4354-a3e0-cb370221cd95",{"tags":34,"relatedLang":11,"relatedPosts":45},[35,37,39,41,43],{"name":20,"slug":36},"build-2026",{"name":17,"slug":38},"microsoft",{"name":18,"slug":40},"mai-models",{"name":21,"slug":42},"llm",{"name":19,"slug":44},"copilot",[46,52,58,64,70,76],{"id":47,"slug":48,"title":49,"cover_image":50,"image_url":50,"created_at":51,"category":13},"f26334ab-dd8b-49c2-a49e-7fc376200f2b","microsoft-bets-on-controllable-domain-tuned-models-zh","微軟押注可控、領域調校模型，而不是更大的通用模型","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781331468190-ymfp.png","2026-06-13T06:17:20.311904+00:00",{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"eaafd0fe-cb56-40cd-80f8-c203e3d72f03","gpt-5-4-thinkng-pro-mini-nano-release-zh","GPT-5.4 率先登場，mini、nano 跟進","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781326973205-xufk.png","2026-06-13T05:02:19.576989+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"09fe28b5-aae5-4bac-b3bd-9a261e4c99a1","mimo-v2-flash-openrouter-benchmarks-pricing-zh","MiMo-V2-Flash 直衝開源 SWE-bench","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781321565467-96el.png","2026-06-13T03:32:17.367685+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"0a9dbc64-2e51-494d-b6b6-21ecfd8dd1f5","minimax-m3-1m-token-coding-power-zh","MiniMax M3 把 1M Token 送進寫碼場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781295477857-78hl.png","2026-06-12T20:17:28.037784+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"8ca34749-2efa-4b24-b2bd-c0fb66062b49","openai-confidential-ipo-us-stock-market-zh","OpenAI 低調送件 IPO","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781294579049-ltqt.png","2026-06-12T20:02:33.419196+00:00",{"id":77,"slug":78,"title":79,"cover_image":80,"image_url":80,"created_at":81,"category":13},"b15b0887-bd5b-43e6-ac42-23939d0f4e92","google-gemini-35-pro-june-2m-token-launch-zh","Gemini 3.5 Pro 6月登場，2M Token 夠猛","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781204585839-bdsh.png","2026-06-11T19:02:36.371587+00:00",[83,88,93,98,103,108,113,118,123,128],{"id":84,"slug":85,"title":86,"created_at":87},"58b64033-7eb6-49b9-9aab-01cf8ae1b2f2","nvidia-rubin-six-chips-one-ai-supercomputer-zh","NVIDIA Rubin 把六顆晶片塞進 AI 機櫃","2026-03-26T07:18:45.861277+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"0dcc2c61-c2a6-480d-adb8-dd225fc68914","march-2026-ai-model-news-what-mattered-zh","2026 年 3 月 AI 模型新聞重點","2026-03-26T07:32:08.386348+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"214ab08b-5ce5-4b5c-8b72-47619d8675dd","why-small-models-are-winning-on-device-ai-zh","小模型為何吃下裝置端 AI","2026-03-26T07:36:30.488966+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"785624b2-0355-4b82-adc3-de5e45eecd88","midjourney-v8-faster-images-higher-costs-zh","Midjourney V8 變快了，也變貴了","2026-03-26T07:52:03.562971+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"cda76b92-d209-4134-86c1-a60f5bc7b128","xiaomi-mimo-trio-agents-robots-voice-zh","小米 MiMo 三模型瞄準代理、機器人與語音","2026-03-28T03:05:08.779489+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"9e1044b4-946d-47fe-9e2a-c2ee032e1164","xiaomi-mimo-v2-pro-1t-moe-agents-zh","小米 MiMo-V2-Pro 登場：1T MoE 模型","2026-03-28T03:06:19.002353+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"c4b6186f-bd84-4598-997e-c6e31d543c0d","cursor-composer-2-agentic-coding-model-zh","Cursor Composer 2 走向代理式寫碼","2026-03-28T03:13:06.422716+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"e112e76f-ec3b-408f-810e-e93ae21a888a","apple-siri-gemini-distilled-models-zh","Apple Siri 牽手 Gemini 的真相","2026-03-29T04:52:57.886544+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"c679b51f-194a-463b-87fc-7695256ff752","mimo-v2-pro-vs-omni-vs-flash-2026-zh","MiMo V2 Pro、Omni、Flash 怎麼選","2026-04-02T01:18:43.576128+00:00",{"id":129,"slug":130,"title":131,"created_at":132},"3b988fd7-6749-4f01-ba25-c0ad7486dc31","z-ai-glm-5v-turbo-design2code-claude-zh","GLM-5V-Turbo 在 Design2Code 贏了…","2026-04-02T04:03:36.31741+00:00"]