[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-apples-foundation-models-are-all-apple-zh":3,"article-related-apples-foundation-models-are-all-apple-zh":32,"series-model-release-5501c878-b81a-4c12-8df6-7876ba68ff43":85},{"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":24,"views":28,"created_at":29,"published_at":30,"topic_cluster_id":31},"5501c878-b81a-4c12-8df6-7876ba68ff43","apples-foundation-models-are-all-apple-zh","Apple Foundation Models 不是 Gemini 包裝","\u003Cp data-speakable=\"summary\">Apple 的新 Foundation Models 由 Apple 程式碼與架構主導，\u003Ca href=\"\u002Ftag\u002Fgoogle\">Google\u003C\u002Fa> 只參與\u003Ca href=\"\u002Fnews\u002Fllm-research-engineers-post-training-services-zh\">訓練\u003C\u002Fa>協助，正式上線的模型仍是 Apple 軟體。\u003C\u002Fp>\u003Cp>說真的，這件事比傳聞冷靜很多。Apple 在 WWDC 2026 把話講白了，不是 \u003Ca href=\"\u002Ftag\u002Fgemini\">Gemini\u003C\u002Fa> 外掛，也不是把 Google 模型包一層皮就上架。\u003C\u002Fp>\u003Cp>它公開了 4 組模型家族，還有 \u003Ca href=\"https:\u002F\u002Fwww.apple.com\u002Fapple-intelligence\u002F\" target=\"_blank\" rel=\"noopener\">Apple Intelligence\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fsupport.apple.com\u002Fen-us\u002Fguide\u002Fsecurity\u002Fsec59b0b31ff\u002Fweb\" target=\"_blank\" rel=\"noopener\">Private Cloud Compute\u003C\u002Fa> 這條雲端路徑。\u003Ca href=\"\u002Fnews\u002Fjensen-huang-lg-ai-cooperation-five-bets-zh\">重點\u003C\u002Fa>很簡單：訓練可以借力，產品體驗一定要自己掌控。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>模型\u003C\u002Fth>\u003Cth>執行位置\u003C\u002Fth>\u003Cth>用途\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>AFM Core\u003C\u002Ftd>\u003Ctd>裝置端\u003C\u002Ftd>\u003Ctd>本地任務的基礎模型\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM Core Advanced\u003C\u002Ftd>\u003Ctd>裝置端\u003C\u002Ftd>\u003Ctd>多模態、稀疏架構模型\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM Cloud\u003C\u002Ftd>\u003Ctd>Private Cloud Compute\u003C\u002Ftd>\u003Ctd>處理裝置端吃不下的請求\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM Cloud Image\u003C\u002Ftd>\u003Ctd>Private Cloud Compute\u003C\u002Ftd>\u003Ctd>影像生成與編修\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM Cloud Pro\u003C\u002Ftd>\u003Ctd>Private Cloud Compute 與 Google 雲端伺服器、NVIDIA GPU\u003C\u002Ftd>\u003Ctd>代理任務與最重工作\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Apple 這次講得很直白\u003C\u002Fh2>\u003Cp>Apple 這次的說法，其實是在拆掉一個很大的誤會。外界原本以為，它可能把 Gemini 直接包進 Siri AI 或 \u003Ca href=\"\u002Ftag\u002Fapple-intelligence\">Apple Intelligence\u003C\u002Fa>。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781404366205-7zqq.png\" alt=\"Apple Foundation Models 不是 Gemini 包裝\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>但 Apple 的說法不是這樣。它說模型是為 Apple Silicon 客製，訓練時用了專有\u003Ca href=\"\u002Fnews\u002Fllm-wiki-compiler-raw-sources-to-wiki-zh\">資料\u003C\u002Fa>，最後部署的是 Apple 自己的軟體堆疊。\u003C\u002Fp>\u003Cp>這差很多。前者像是租別人的引擎來開車，後者比較像自己做引擎，但會參考別人的賽道資料。\u003C\u002Fp>\u003Cp>Apple 也把系統拆得很清楚。小任務留在裝置端，重任務送到 Private Cloud Compute，最吃資源的工作再走另一條雲端路徑。\u003C\u002Fp>\u003Cul>\u003Cli>AFM Core 和 AFM Core Advanced 跑在裝置端。\u003C\u002Fli>\u003Cli>AFM Cloud 和 AFM Cloud Image 跑在 Private Cloud Compute。\u003C\u002Fli>\u003Cli>AFM Cloud Pro 會用 Google 雲端伺服器與 NVIDIA GPU。\u003C\u002Fli>\u003Cli>Apple 強調使用者接觸到的仍是 Apple 軟體。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>為什麼大家會把它聽成 Gemini\u003C\u002Fh2>\u003Cp>原因其實不難懂。Apple 之前提過 Google 相關技術，這種說法很容易讓人腦補成「Apple 直接投降了」。\u003C\u002Fp>\u003Cp>但從它現在的描述看來，Google 的角色比較像訓練協助。真正跑在使用者面前的模型、伺服器管理、產品介面，還是 Apple 自己握著。\u003C\u002Fp>\u003Cp>這種切法很 Apple。它很少把核心體驗交給別人，尤其是 AI 這種會碰到隱私、延遲、成本的東西。\u003C\u002Fp>\u003Cblockquote>“We use none of the models that Google deploys to their customers, nor do we use the infrastructure and means by which they deploy models to their customers.” — Apple executive during WWDC 2026 discussion, as quoted in AppleInsider\u003C\u002Fblockquote>\u003Cp>這句話很重。意思很清楚：Apple 要的是技術支援，不是把產品靈魂交出去。\u003C\u002Fp>\u003Cp>對開發者來說，這也代表一件事。Apple 的 AI API 和模型策略，會繼續跟 Google、\u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa>、\u003Ca href=\"\u002Ftag\u002Fanthropic\">Anthropic\u003C\u002Fa> 走不同路線。\u003C\u002Fp>\u003Ch2>Apple 其實是在補 2024 之後的洞\u003C\u002Fh2>\u003Cp>Apple Intelligence 一開始就讓很多人失望。功能延後、節奏不穩，大家自然會懷疑它是不是跟不上其他 LLM 產品。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781404367254-riih.png\" alt=\"Apple Foundation Models 不是 Gemini 包裝\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>所以這次只要一提到 Google，外界就會立刻放大解讀。這不是沒原因，是 Apple 自己前一輪 AI 交作業沒交漂亮。\u003C\u002Fp>\u003Cp>但從架構看，Apple 這次明顯更務實。把推論留在裝置端，可以少碰使用者資料；把大任務放進 Private Cloud Compute，可以保住隱私敘事；再把最重的工作丟給雲端，成本也比較好控。\u003C\u002Fp>\u003Cul>\u003Cli>裝置端推論，資料外流風險更低。\u003C\u002Fli>\u003Cli>Private Cloud Compute 讓 Apple 有中間層。\u003C\u002Fli>\u003Cli>Cloud Pro 提供更高算力。\u003C\u002Fli>\u003Cli>Distillation 讓 Apple 吸收模型知識，不必直接上架原模型。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這套做法跟其他 AI 廠商差在哪\u003C\u002Fh2>\u003Cp>如果拿 \u003Ca href=\"https:\u002F\u002Fopenai.com\" target=\"_blank\" rel=\"noopener\">OpenAI\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\" target=\"_blank\" rel=\"noopener\">Anthropic\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fdeepmind.google\" target=\"_blank\" rel=\"noopener\">Google DeepMind\u003C\u002Fa> 來看，Apple 的路線很不一樣。別人多半先做一個很強的通用模型，再想辦法塞進產品。\u003C\u002Fp>\u003Cp>Apple 則是反過來。它先定義裝置端、雲端、隱私、延遲，再回頭設計模型。這種做法很慢，但比較符合它的硬體生態。\u003C\u002Fp>\u003Cp>講白了，Apple 不想做成一個純 API 公司。它要的是模型、晶片、系統、雲端一起配合，讓 AI 看起來像原生功能，而不是外掛服務。\u003C\u002Fp>\u003Cp>這裡可以直接列一下差異。\u003C\u002Fp>\u003Cul>\u003Cli>OpenAI 偏向模型能力優先。\u003C\u002Fli>\u003Cli>Anthropic 偏向安全與代理工作流。\u003C\u002Fli>\u003Cli>Google DeepMind 偏向大規模研究與產品整合。\u003C\u002Fli>\u003Cli>Apple 偏向裝置端體驗與控制權。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這對 Siri 和 Apple Intelligence 代表什麼\u003C\u002Fh2>\u003Cp>最直接的答案是，Siri 終於有機會換一顆比較像樣的腦袋。Apple 把新 Foundation Models 接到 Siri AI 和 Apple Intelligence，上面那層體驗才是大家真正會碰到的東西。\u003C\u002Fp>\u003Cp>如果它做得好，使用者感受到的會是更快的回應、更穩的上下文理解，還有更少的雲端等待時間。這些都很無聊，但很重要。\u003C\u002Fp>\u003Cp>如果它做不好，那就還是老問題。演示很漂亮，實際使用卻卡卡的，然後大家又開始拿 iPhone 跟 Android 陣營比 AI。\u003C\u002Fp>\u003Cp>Apple 現在的優勢，是它有完整硬體鏈。A 系列與 M 系列晶片、作業系統、雲端層都在自己手上，這讓它可以玩出別家很難複製的本地推論策略。\u003C\u002Fp>\u003Cp>但弱點也很明顯。它不能只靠品牌撐場面。模型品質、回應速度、工具調用能力，還有多輪對話穩定性，最後都要拿真機驗證。\u003C\u002Fp>\u003Ch2>這場爭議其實在講 AI 控制權\u003C\u002Fh2>\u003Cp>這次事件不是單純的技術八卦。它在講一個更大的問題：AI 到底是誰在控制。\u003C\u002Fp>\u003Cp>如果一家公司把模型、資料、伺服器、介面全交給別人，那它很快就會變成轉接頭。短期省事，長期很難守住產品差異。\u003C\u002Fp>\u003Cp>Apple 顯然不想走那條路。它可以借 Google 的訓練方法，也可以用外部雲端資源，但它要保留最後的決定權。\u003C\u002Fp>\u003Cp>我覺得這才是重點。不是 Apple 有沒有用 Google，而是 Apple 有沒有讓 Google 進到使用者看得到的那一層。\u003C\u002Fp>\u003Cp>目前答案看起來是沒有。接下來就看它能不能把這套架構真的跑順，別再讓 Siri 變成笑話。\u003C\u002Fp>\u003Cp>如果 Apple 真的能把裝置端模型、Private Cloud Compute、Cloud Pro 串成穩定體驗，那它在 2026 之後的 AI 競爭裡，會很難再被說成只會慢半拍。反過來說，只要使用者一測就破功，這套說法也撐不久。\u003C\u002Fp>","Apple 在 WWDC 2026 說明，新 Foundation Models 由 Apple 程式碼與架構主導，Google 只參與訓練協助，正式上線的模型仍是 Apple 軟體。","appleinsider.com","https:\u002F\u002Fappleinsider.com\u002Farticles\u002F26\u002F06\u002F08\u002Fapples-new-foundation-models-dont-contain-a-drop-of-gemini-as-we-said-they-wouldnt",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781404366205-7zqq.png","model-release","zh","9014154a-46b0-4613-9545-a87c02665870",[17,18,19,20,21,22,23],"Apple Foundation Models","Apple Intelligence","Siri AI","Private Cloud Compute","WWDC 2026","Google Gemini","LLM",[25,26,27],"Apple 這次不是把 Gemini 直接包裝成自家模型。","Google 參與的是訓練協助，不是使用者端產品堆疊。","Apple 把模型分成裝置端、私有雲端與高算力雲端三層。",0,"2026-06-14T02:32:25.319584+00:00","2026-06-14T02:32:25.311+00:00","0a3b4f35-7be1-430e-b708-37bdc8b5219a",{"tags":33,"relatedLang":44,"relatedPosts":48},[34,36,38,40,42],{"name":19,"slug":35},"siri-ai",{"name":21,"slug":37},"wwdc-2026",{"name":18,"slug":39},"apple-intelligence",{"name":20,"slug":41},"private-cloud-compute",{"name":17,"slug":43},"apple-foundation-models",{"id":15,"slug":45,"title":46,"language":47},"apples-foundation-models-are-all-apple-en","Apple’s new Foundation Models are all Apple","en",[49,55,61,67,73,79],{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"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":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":13},"0bb91791-f4b6-4d51-899c-6eeb239f942a","microsoft-mai-models-build-2026-zh","Microsoft把 Copilot 拉回主場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781330585064-s9ya.png","2026-06-13T06:02:35.160901+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"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":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"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":74,"slug":75,"title":76,"cover_image":77,"image_url":77,"created_at":78,"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":80,"slug":81,"title":82,"cover_image":83,"image_url":83,"created_at":84,"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",[86,91,96,101,106,111,116,121,126,131],{"id":87,"slug":88,"title":89,"created_at":90},"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":92,"slug":93,"title":94,"created_at":95},"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":97,"slug":98,"title":99,"created_at":100},"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":102,"slug":103,"title":104,"created_at":105},"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":107,"slug":108,"title":109,"created_at":110},"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":112,"slug":113,"title":114,"created_at":115},"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":117,"slug":118,"title":119,"created_at":120},"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":122,"slug":123,"title":124,"created_at":125},"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":127,"slug":128,"title":129,"created_at":130},"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":132,"slug":133,"title":134,"created_at":135},"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"]