[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-microsoft-launches-three-in-house-ai-models-zh":3,"tags-microsoft-launches-three-in-house-ai-models-zh":33,"related-lang-microsoft-launches-three-in-house-ai-models-zh":50,"related-posts-microsoft-launches-three-in-house-ai-models-zh":54,"series-model-release-33ef53ea-c20f-49ed-85ad-96e19be90ab6":91},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":21,"translated_content":10,"views":22,"is_premium":23,"created_at":24,"updated_at":24,"cover_image":11,"published_at":25,"rewrite_status":26,"rewrite_error":10,"rewritten_from_id":27,"slug":28,"category":29,"related_article_id":30,"status":31,"google_indexed_at":32,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":10,"topic_cluster_id":10,"embedding":10,"is_canonical_seed":23},"33ef53ea-c20f-49ed-85ad-96e19be90ab6","Microsoft 推出三款自家 AI 模型","\u003Cp>Microsoft 這次不是小修小補。它直接端出三款自家 AI 模型，分別做文字、語音、圖片。時間點也很硬：2026 年 4 月 2 日。\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002F\" target=\"_blank\" rel=\"noopener\">Microsoft\u003C\u002Fa> 這回明講了，自己要在模型層多踩一腳。\u003C\u002Fp>\u003Cp>最吸睛的是 \u003Cstrong>MAI-Transcribe-1\u003C\u002Fstrong>。它支援 25 種語言，還號稱比 Azure Fast 快 2.5 倍。\u003Cstrong>MAI-Voice-1\u003C\u002Fstrong> 更誇張，1 秒能生出 60 秒音訊。這種數字很像在對市場喊話：我不只會做，我還要做得便宜。\u003C\u002Fp>\u003Ch2>Microsoft 這次到底丟了什麼\u003C\u002Fh2>\u003Cp>這三款模型來自 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai\" target=\"_blank\" rel=\"noopener\">Microsoft AI\u003C\u002Fa>。團隊由 Mustafa Suleyman 領軍。名字分別是 \u003Cstrong>MAI-Transcribe-1\u003C\u002Fstrong>、\u003Cstrong>MAI-Voice-1\u003C\u002Fstrong>、\u003Cstrong>MAI-Image-2\u003C\u002Fstrong>。它們會先進到 \u003Ca href=\"https:\u002F\u002Fazure.microsoft.com\u002Fproducts\u002Fai-foundry\" target=\"_blank\" rel=\"noopener\">Microsoft Foundry\u003C\u002Fa> 和 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai\u002Fai-playground\" target=\"_blank\" rel=\"noopener\">MAI Playground\u003C\u002Fa>，讓開發者先試，再決定要不要上線。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775172105180-ufot.png\" alt=\"Microsoft 推出三款自家 AI 模型\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>講白了，\u003Ca href=\"\u002Fnews\u002Fmicrosoft-adds-multi-model-copilot-workflows-zh\">Micr\u003C\u002Fa>osoft 不是只想賣雲端。它想把模型層也握在手上。這很符合它一貫套路：先把入口做大，再把工具鏈黏住你。對企業客戶來說，這種做法很實際，因為採購、測試、部署可以放在同一套流程裡。\u003C\u002Fp>\u003Cp>而且這次不是單打獨鬥。\u003Ca href=\"\u002Fnews\u002Fmidjourney-public-beta-visual-generation-history-zh\">Mi\u003C\u002Fa>crosoft 一邊跟 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002F\" target=\"_blank\" rel=\"noopener\">OpenAI\u003C\u002Fa> 綁很深，一邊又自己養模型。這種雙線打法，說穿了就是保留退路。AI 市場變化太快，誰都不想把命運全押在別人身上。\u003C\u002Fp>\u003Cul>\u003Cli>MAI-Transcribe-1 支援 25 種語言\u003C\u002Fli>\u003Cli>Microsoft 說它比 Azure Fast 快 2.5 倍\u003C\u002Fli>\u003Cli>MAI-Voice-1 可在 1 秒內生出 60 秒音訊\u003C\u002Fli>\u003Cli>MAI-Transcribe-1 起價是每小時 0.36 美元\u003C\u002Fli>\u003Cli>MAI-Voice-1 起價是每 100 萬字元 22 美元\u003C\u002Fli>\u003Cli>MAI-Image-2 起價是每 100 萬文字輸入 token 5 美元\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這些價格很有意思。它們不是在拚一個超大通用模型。它們是在拚任務單價。對企業來說，這才是重點。你做客服轉寫、語音助理、內容生成，成本只要少一點，財務部就會比較好看。\u003C\u002Fp>\u003Ch2>Suleyman 想把 AI 說成更像工具\u003C\u002Fh2>\u003Cp>這批模型由 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai\u002Fai-superintelligence\" target=\"_blank\" rel=\"noopener\">MAI Superintelligence team\u003C\u002Fa> 做出來。這個團隊在 2025 年 11 月成立。Suleyman 在官方文章裡提到一個詞：Human\u003Ca href=\"\u002Fnews\u002Fmistral-voxtral-tts-open-source-voice-ai-zh\">ist\u003C\u002Fa> AI。聽起來很文青，但意思其實很直白，就是把人放在中心，模型要更貼近真實溝通。\u003C\u002Fp>\u003Cp>他原話是這樣說的：\u003C\u002Fp>\u003Cblockquote>“At Microsoft AI, we’re building Humanist AI. We have a distinct view when creating our AI models — putting humans at the center, optimizing for how people actually communicate, training for practical use.”\u003C\u002Fblockquote>\u003Cp>這段話不是只有公關味。它也在講產品方向。Microsoft 想把模型做成能直接進工作流程的工具。不是只拿來跑 benchmark，也不是只在 demo 場合帥一下。對台灣開發者來說，這種定位很熟悉，就是先求穩，再求快。\u003C\u002Fp>\u003Cp>Suleyman 也一直在處理和 OpenAI 的關係。他在接受 \u003Ca href=\"https:\u002F\u002Fventurebeat.com\u002F\" target=\"_blank\" rel=\"noopener\">VentureBeat\u003C\u002Fa> 訪問時，還是強調合作關係沒斷。\u003Ca href=\"https:\u002F\u002Fwww.theverge.com\u002F\" target=\"_blank\" rel=\"noopener\">The Verge\u003C\u002Fa> 也提到，新的協議讓 Microsoft 在自家超級智慧研究上有更多空間。這代表什麼？代表它不想只當大金主，它想當自己模型路線的主控方。\u003C\u002Fp>\u003Ch2>數字很直接，目標也很直接\u003C\u002Fh2>\u003Cp>AI 市場現在很擠。\u003Ca href=\"https:\u002F\u002Fcloud.google.com\u002Fvertex-ai\u002Fgenerative-ai\" target=\"_blank\" rel=\"noopener\">Google Vertex AI\u003C\u002Fa>、OpenAI、Anthropic，還有一堆新創，都在搶同一批預算。這時候比誰更會講故事，意義沒那麼大。真正有用的是：誰的 API 夠快、夠便宜、夠穩。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775172109746-slxg.png\" alt=\"Microsoft 推出三款自家 AI 模型\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Microsoft 這次給的數字，幾乎就是在打這個點。MAI-Transcribe-1 每小時 0.36 美元。MAI-Voice-1 每 100 萬字元 22 美元。MAI-Image-2 則是每 100 萬文字輸入 token 5 美元、每 100 萬圖片輸出 token 33 美元。這些都很像在說：我不是來秀肌肉，我是來搶工作量。\u003C\u002Fp>\u003Cp>如果你是開發者，你會很快想到幾個場景。客服錄音轉文字、會議逐字稿、語音導覽、App 內配音、商品圖生成。這些都不是玩具需求。它們是會燒錢的真實需求。只要單次成本降一點，整個產品的毛利就會比較像樣。\u003C\u002Fp>\u003Cp>更重要的是，Microsoft 把這些模型放進 Foundry。這表示企業不用重做整套 AI 管線。你原本就在 Azure 上跑服務，現在多半只要換模型端點。這種低摩擦切換，對採購部門來說很有吸引力。\u003C\u002Fp>\u003Cul>\u003Cli>MAI-Transcribe-1：25 語言，2.5 倍速度優勢\u003C\u002Fli>\u003Cli>MAI-Voice-1：1 秒生成 60 秒音訊\u003C\u002Fli>\u003Cli>MAI-Image-2：同時涵蓋文字輸入與圖片輸出計價\u003C\u002Fli>\u003Cli>OpenAI：強在通用能力與生態聲量\u003C\u002Fli>\u003Cli>Google：強在雲端整合與多模態產品線\u003C\u002Fli>\u003Cli>Microsoft：強在企業合約與 Azure 既有客戶\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這跟競品比，差在哪\u003C\u002Fh2>\u003Cp>如果只看模型能力，大家都會說自己很強。這種話聽多了，真的會膩。比較有意思的是產品路線。OpenAI 比較像把旗艦模型先推到前面，再慢慢往企業產品補齊。Google 則是把模型塞進雲端和搜尋生態。Microsoft 這次的打法很明確：直接把模型變成企業可採購的零件。\u003C\u002Fp>\u003Cp>這種打法的優點很現實。第一，採購流程短。第二，帳單容易算。第三，開發者不必換整套平台。對很多台灣公司來說，這比模型排行榜上的 0.5 分差距更重要。因為真的上線後，大家只看成本和 SLA。\u003C\u002Fp>\u003Cp>如果拿語音轉寫來看，MAI-Transcribe-1 的賣點很清楚。25 種語言夠用，2.5 倍速度也夠有感。對客服中心、媒體、教育平台來說，這種速度差異會直接反映在等待時間和伺服器成本上。這不是抽象優勢，是帳單上的差別。\u003C\u002Fp>\u003Cp>再看語音生成。\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai\" target=\"_blank\" rel=\"noopener\">Microsoft AI\u003C\u002Fa> 說 MAI-Voice-1 可以 1 秒產生 60 秒音訊。這個速度很猛，但實務上還要看音質、情緒控制、停頓自然度。因為企業不是只要快，還要能用。這就是和 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002F\" target=\"_blank\" rel=\"noopener\">OpenAI\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fcloud.google.com\u002Fvertex-ai\u002Fgenerative-ai\" target=\"_blank\" rel=\"noopener\">Google\u003C\u002Fa> 拉開差距的地方。\u003C\u002Fp>\u003Ch2>背後其實是平台戰\u003C\u002Fh2>\u003Cp>Microsoft 這步棋，表面上是推出三個模型。實際上是把平台戰再往前推一格。它想讓開發者在 Azure 裡就能完成測試、部署、計費。這種整合式路線，對大型企業很有用，因為 IT 團隊最怕東拼西湊。\u003C\u002Fp>\u003Cp>另一個背景是成本壓力。現在很多公司都在算 AI ROI。模型再強，若每次呼叫都貴到爆，最後還是只能放在 demo。Microsoft 這次把價格壓到一個明顯能談商務的區間，就是要讓 AI 變成可大量使用的基礎設施，而不是只給產品經理拍片用。\u003C\u002Fp>\u003Cp>我覺得這也反映一件事。AI 競爭早就不是單純比誰模型大。現在比的是誰能把模型塞進既有雲端、資料、權限、付款和監控系統。誰能少讓工程師重寫一段 code，誰就比較容易拿到合約。\u003C\u002Fp>\u003Cp>對台灣開發團隊來說，這類消息的重點很簡單。不要只盯著模型名稱。要看價格表、延遲、區域支援、語言品質、以及能不能直接接你現在的 API 架構。說真的，這些才是會決定你要不要切換的東西。\u003C\u002Fp>\u003Ch2>接下來要看什麼\u003C\u002Fh2>\u003Cp>我會先看兩件事。第一，這三款模型會不會很快進到更多 Microsoft 產品。第二，實際第三方測試會不會跟官方數字差很多。AI 圈很常見這種狀況：簡報很漂亮，上線後才知道真相。\u003C\u002Fp>\u003Cp>如果 MAI-Transcribe-1 和 MAI-Voice-1 的表現真的穩，最先受惠的會是客服、媒體、教育和企業內部工具。這些場景對速度和成本都很敏感。反過來說，如果品質不夠穩，市場也不會客氣，因為企業客戶最會算帳。\u003C\u002Fp>\u003Cp>我的預測很直接。Microsoft 接下來會把更多自家模型塞進 Foundry，然後用價格和整合度去搶企業單。你如果是開發者，現在就該試試看這些 API。不要等別人先把流程接完，才回頭補功課。\u003Ca href=\"https:\u002F\u002Fazure.microsoft.com\u002Fproducts\u002Fai-foundry\" target=\"_blank\" rel=\"noopener\">Microsoft Foundry\u003C\u002Fa> 和 \u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai\u002Fai-playground\" target=\"_blank\" rel=\"noopener\">MAI Playground\u003C\u002Fa> 已經開著了，直接玩最準。\u003C\u002Fp>","Microsoft 一口氣推出文字、語音、圖片三款自家 AI 模型。MAI-Transcribe-1 主打 25 種語言、速度比 Azure Fast 快 2.5 倍，價格也更低，直接把企業採購壓力丟回市場。","techcrunch.com","https:\u002F\u002Ftechcrunch.com\u002F2026\u002F04\u002F02\u002Fmicrosoft-takes-on-ai-rivals-with-three-new-foundational-models\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775172105180-ufot.png",[13,14,15,16,17,18,19,20],"Microsoft","AI模型","MAI-Transcribe-1","MAI-Voice-1","MAI-Image-2","Azure","Foundry","OpenAI","zh",1,false,"2026-04-02T23:21:30.619885+00:00","2026-04-02T23:21:30.534+00:00","done","0ae8aef3-b22f-4deb-8a52-7b691cfeb1c5","microsoft-launches-three-in-house-ai-models-zh","model-release","78908214-aba5-4673-b3ae-cf432c1826c9","published","2026-04-07T07:41:14.411+00:00",[34,36,38,40,42,44,46,48],{"name":15,"slug":35},"mai-transcribe-1",{"name":13,"slug":37},"microsoft",{"name":20,"slug":39},"openai",{"name":17,"slug":41},"mai-image-2",{"name":14,"slug":43},"ai模型",{"name":18,"slug":45},"azure",{"name":16,"slug":47},"mai-voice-1",{"name":19,"slug":49},"foundry",{"id":30,"slug":51,"title":52,"language":53},"microsoft-launches-three-in-house-ai-models-en","Microsoft launches three in-house AI models","en",[55,61,67,73,79,85],{"id":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":29},"bd8cfc0e-66db-4546-9b9e-fa328f7538d6","weishenme-google-yincang-de-gemini-live-moxing-bi-yanshi-gen-zh","為什麼 Google 隱藏的 Gemini Live 模型，比演示更重要","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778869245574-c25w.png","2026-05-15T18:20:23.111559+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"category":29},"5b5fa24f-5259-4e9e-8270-b08b6805f281","minimax-m1-open-hybrid-attention-reasoning-model-zh","MiniMax-M1：開源 1M Token 推理模型","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778797859209-ea1g.png","2026-05-14T22:30:38.636592+00:00",{"id":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"category":29},"b1da56ac-8019-4c6b-a8dc-22e6e22b1cb5","gemini-omni-video-review-text-rendering-zh","Gemini Omni 影片模型怎麼了","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778779280109-lrrk.png","2026-05-14T17:20:42.608312+00:00",{"id":74,"slug":75,"title":76,"cover_image":77,"image_url":77,"created_at":78,"category":29},"d63e9d93-e613-4bbf-8135-9599fde11d08","why-xiaomi-mimo-v25-pro-changes-coding-agents-zh","為什麼 Xiaomi 的 MiMo-V2.5-Pro 改變的是 Coding …","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778689858139-v38e.png","2026-05-13T16:30:27.893951+00:00",{"id":80,"slug":81,"title":82,"cover_image":83,"image_url":83,"created_at":84,"category":29},"8f0c9185-52f9-46f2-82c6-5baec126ba2e","openai-realtime-audio-models-live-voice-zh","OpenAI 即時音訊模型瞄準語音互動","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778451657895-2iu7.png","2026-05-10T22:20:32.443798+00:00",{"id":86,"slug":87,"title":88,"cover_image":89,"image_url":89,"created_at":90,"category":29},"52106dc2-4eba-4ca0-8318-fa646064de97","anthropic-10-finance-ai-agents-zh","Anthropic推10款金融AI Agent","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778389843399-vclb.png","2026-05-10T05:10:22.778762+00:00",[92,97,102,107,112,117,122,127,132,137],{"id":93,"slug":94,"title":95,"created_at":96},"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":98,"slug":99,"title":100,"created_at":101},"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":103,"slug":104,"title":105,"created_at":106},"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":108,"slug":109,"title":110,"created_at":111},"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":113,"slug":114,"title":115,"created_at":116},"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":118,"slug":119,"title":120,"created_at":121},"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":123,"slug":124,"title":125,"created_at":126},"d68e59a2-55eb-4a8f-95d6-edc8fcbff581","cursor-composer-2-started-from-kimi-zh","Cursor Composer 2 其實從 Kimi 起步","2026-03-28T03:11:58.893796+00:00",{"id":128,"slug":129,"title":130,"created_at":131},"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":133,"slug":134,"title":135,"created_at":136},"45812c46-99fc-4b1f-aae1-56f64f5c9024","openai-shuts-down-sora-video-app-api-zh","OpenAI 關閉 Sora App 與 API","2026-03-29T04:47:48.974108+00:00",{"id":138,"slug":139,"title":140,"created_at":141},"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"]