[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-apple-gemini-backed-ai-still-its-own-thing-en":3,"article-related-apple-gemini-backed-ai-still-its-own-thing-en":30,"series-model-release-ee040a31-a4bc-41fb-8430-e6911d9937de":79},{"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":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"ee040a31-a4bc-41fb-8430-e6911d9937de","apple-gemini-backed-ai-still-its-own-thing-en","Apple’s Gemini-backed AI is still its own thing","\u003Cp data-speakable=\"summary\">Apple rebuilt \u003Ca href=\"\u002Ftag\u002Fgoogle\">Google\u003C\u002Fa>-derived \u003Ca href=\"\u002Ftag\u002Fgemini\">Gemini\u003C\u002Fa> models into \u003Ca href=\"\u002Ftag\u002Fapple-intelligence\">Apple Intelligence\u003C\u002Fa> with its own data and guardrails.\u003C\u002Fp>\u003Cp>\u003Ca href=\"\u002Ftag\u002Fapple\">Apple\u003C\u002Fa>’s WWDC26 AI update is messy in exactly the way real infrastructure is messy: part on-device, part cloud, and one model still running on \u003Ca href=\"https:\u002F\u002Fcloud.google.com\" target=\"_blank\" rel=\"noopener\">Google Cloud\u003C\u002Fa>. The key detail is that Apple says the system is no longer a simple Google skin, even though it started from Google code.\u003C\u002Fp>\u003Cp>That matters because Apple is trying to sell a very specific story about privacy, performance, and control. The company now says its third generation of Apple Foundation Models includes five models, with two running locally and three in the cloud, including one tied to Google’s servers.\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Model\u003C\u002Fth>\u003Cth>Where it runs\u003C\u002Fth>\u003Cth>What it does\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>AFM 3 Core\u003C\u002Ftd>\u003Ctd>On-device\u003C\u002Ftd>\u003Ctd>Siri smarts\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM 3 Core Advanced\u003C\u002Ftd>\u003Ctd>On-device\u003C\u002Ftd>\u003Ctd>More expressive voices, better dictation\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM 3 Cloud\u003C\u002Ftd>\u003Ctd>Cloud\u003C\u002Ftd>\u003Ctd>General server-side intelligence\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>ADM 3 Cloud\u003C\u002Ftd>\u003Ctd>Cloud\u003C\u002Ftd>\u003Ctd>Image generation and editing\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>AFM 3 Cloud Pro\u003C\u002Ftd>\u003Ctd>Google Cloud\u003C\u002Ftd>\u003Ctd>Party-planning and higher-end cloud tasks\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Apple’s AI stack is split across device and cloud\u003C\u002Fh2>\u003Cp>Apple’s new setup is built around the idea that different AI jobs need different hardware. The lighter tasks stay on the device, which helps with latency and privacy. The heavier tasks move to the cloud, where Apple can use more compute without asking every iPhone to pretend it is a data center.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781706790424-qfgw.png\" alt=\"Apple’s Gemini-backed AI is still its own thing\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>The article from \u003Ca href=\"https:\u002F\u002Fwww.macworld.com\" target=\"_blank\" rel=\"noopener\">Macworld\u003C\u002Fa> points to five models in total. Two are local models: AFM 3 Core and AFM 3 Core Advanced. Three are cloud models: AFM 3 Cloud, ADM 3 Cloud, and AFM 3 Cloud Pro.\u003C\u002Fp>\u003Cul>\u003Cli>AFM 3 Core handles Siri improvements on-device.\u003C\u002Fli>\u003Cli>AFM 3 Core Advanced adds more expressive voices and stronger dictation.\u003C\u002Fli>\u003Cli>AFM 3 Cloud Pro is the only model running on Google’s servers.\u003C\u002Fli>\u003Cli>Apple says the models were rebuilt for Apple Silicon and Apple’s device sizes.\u003C\u002Fli>\u003C\u002Ful>\u003Cp>That split is the real story. Apple is not shipping a single giant model and calling it a day. It is slicing the work into layers, then placing each layer where Apple thinks it makes the most sense technically and commercially.\u003C\u002Fp>\u003Cp>Jason Cross of Macworld says Apple started with Gemini’s foundation models, then optimized and rebuilt them for Apple Silicon, retrained them with Apple data, weights, and guardrails. That is a much more specific claim than “Apple uses Gemini,” and it changes how you should read the whole thing.\u003C\u002Fp>\u003Ch2>Apple says it rebuilt the models, not just renamed them\u003C\u002Fh2>\u003Cp>The distinction between “derived from Google code” and “Google’s AI in Apple clothing” is doing a lot of work here. Apple wants credit for the parts that matter to users: the training data, the weights, the guardrails, and the way the models behave on its hardware.\u003C\u002Fp>\u003Cblockquote>“It seems like Apple started with Gemini’s foundation models, optimized and rebuilt them for Apple Silicon and the model sizes it needs, and retrained them with its own data, weights, and guardrails.” — Jason Cross, Macworld\u003C\u002Fblockquote>\u003Cp>That quote gets to the heart of the issue. In AI, the starting point matters less than the final tuning. A model can begin life in one lab and end up behaving very differently after retraining, safety filtering, and hardware-specific changes.\u003C\u002Fp>\u003Cp>Apple’s approach also fits its broader strategy. The company has spent years making the case that it can control the full stack better than rivals can. Here, that means keeping some intelligence on-device, pushing more demanding tasks to the cloud, and limiting how much of the plumbing depends on outside infrastructure.\u003C\u002Fp>\u003Ch2>The numbers show how hybrid this system really is\u003C\u002Fh2>\u003Cp>The technical details make Apple’s position less tidy than the marketing might suggest. But they also make the system more interesting, because Apple is clearly choosing pragmatism over purity.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781706796490-3p2x.png\" alt=\"Apple’s Gemini-backed AI is still its own thing\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cul>\u003Cli>5 total Apple Foundation Models in the third generation.\u003C\u002Fli>\u003Cli>2 models run locally on Apple devices.\u003C\u002Fli>\u003Cli>3 models run in the cloud.\u003C\u002Fli>\u003Cli>1 model runs on Google’s servers.\u003C\u002Fli>\u003Cli>WWDC26 was the moment Apple disclosed the setup.\u003C\u002Fli>\u003C\u002Ful>\u003Cp>That last point matters because the confusion before WWDC26 was not accidental. If Apple had been clearer earlier, the debate would have been simpler. Instead, the company let people assume the worst or the laziest explanation: that Apple Intelligence was just Google AI with a fresh coat of paint.\u003C\u002Fp>\u003Cp>It is more complicated than that. Apple appears to have taken Google-originated model work, then reshaped it for its own devices, its own constraints, and its own product goals. That does not erase Google from the story, but it does make the relationship less direct than the headlines imply.\u003C\u002Fp>\u003Ch2>What this means for Apple Intelligence users\u003C\u002Fh2>\u003Cp>For users, the practical question is whether the AI feels fast, private, and useful. If AFM 3 Core and AFM 3 Core Advanced do what Apple claims, the on-device pieces should keep common tasks responsive. The cloud models can handle the heavier lifting without forcing every request through the same path.\u003C\u002Fp>\u003Cp>The bigger issue is trust. Apple is asking people to believe that a model lineage matters less than the final implementation. That is a fair argument if the outputs are good and the privacy story holds up. It is a harder sell if users keep hearing that one of the models is still living on Google’s infrastructure.\u003C\u002Fp>\u003Cp>There is also a cultural problem here. “AI” has become a catch-all label for code completion, image generation, data analysis, and synthetic slop. That fuzziness helps companies bundle good tools with bad ones and sell the whole package as inevitable progress. Apple’s version is cleaner than most, but it still sits inside that same overloaded label.\u003C\u002Fp>\u003Cp>The real test is simple: does Apple Intelligence feel like Apple software, or does it feel like a borrowed model wearing an Apple badge? If Apple keeps the on-device pieces fast and the cloud pieces useful, most people will stop caring where the first version of the model came from. If not, the Google connection will keep leaking into the conversation every time Apple talks about AI.\u003C\u002Fp>\u003Cp>My bet is that Apple will keep tightening the split between local and cloud models, because that is the only way this story stays believable. The next question is whether Apple can make that architecture invisible to users, or whether the Google origin of one model keeps surfacing every time Apple explains how its AI works.\u003C\u002Fp>","Apple’s new Apple Intelligence uses Google-derived models, but Apple rebuilt them with its own weights, data, and guardrails.","www.macworld.com","https:\u002F\u002Fwww.macworld.com\u002Farticle\u002F3166224\u002Fai-is-all-kinds-of-complicated.html",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781706790424-qfgw.png","model-release","en","1d05b1ff-9ced-4fde-bc0d-f21e4775c8c8",[17,18,19,20,21],"Apple Intelligence","Gemini","Apple Foundation Models","WWDC26","Google Cloud",[23,24,25],"Apple says its new AI was rebuilt with Apple data, weights, and guardrails.","The third-generation Apple Foundation Models include 5 models, with 2 on-device and 3 in the cloud.","One cloud model still runs on Google Cloud, which keeps the Google connection alive.",0,"2026-06-17T14:32:45.040826+00:00","2026-06-17T14:32:45.032+00:00","4b06c63f-6ae3-4603-90a5-bd7d94fcf1c5",{"tags":31,"relatedLang":38,"relatedPosts":42},[32,34,36],{"name":21,"slug":33},"google-cloud",{"name":17,"slug":35},"apple-intelligence",{"name":18,"slug":37},"gemini",{"id":15,"slug":39,"title":40,"language":41},"apple-gemini-backed-ai-still-its-own-thing-zh","Apple 的 Gemini 血統，還是 Apple 的 AI","zh",[43,49,55,61,67,73],{"id":44,"slug":45,"title":46,"cover_image":47,"image_url":47,"created_at":48,"category":13},"4515e89e-6fbd-4dd8-a5fa-bbd2bcf6425a","fable-5-drew-rare-praise-ai-voices-en","Fable 5 drew rare praise from top AI voices","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781748174800-46t2.png","2026-06-18T02:02:31.145023+00:00",{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"category":13},"a6017fa4-a339-4a83-b086-16a69dbde34d","devin-pricing-june-2026-plans-limits-en","Devin pricing in June 2026: plans, limits, tradeoffs","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781735574969-l960.png","2026-06-17T22:32:28.222997+00:00",{"id":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":13},"ccc46975-50d1-4ece-8fd3-c082bf4858ae","self-host-minimax-m3-gpu-cloud-en","Self-host MiniMax M3 on GPU cloud","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781716680837-ikof.png","2026-06-17T17:17:35.800599+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"category":13},"17a7dc8b-25e4-4993-b0dd-b23733390007","gemma-4-256k-context-open-models-en","Gemma 4 brings 256K context to open models","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781686085345-8dgk.png","2026-06-17T08:47:34.623499+00:00",{"id":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"category":13},"15982ebe-5f2f-44c0-ade7-6f47a149cb1e","kimi-k2-7-code-api-kimi-code-first-en","Kimi K2.7 Code 该优先上 API 和 Kimi Code，而不是等生态成熟","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781631170958-dj9h.png","2026-06-16T17:32:23.025066+00:00",{"id":74,"slug":75,"title":76,"cover_image":77,"image_url":77,"created_at":78,"category":13},"68ef0621-c9fa-4a58-9bdf-51c3f5ac6bce","kingdom-hearts-iv-confirmed-switch-2-launch-en","Kingdom Hearts IV confirmed for Switch 2 launch","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781615872914-vfne.png","2026-06-16T13:17:24.943789+00:00",[80,85,90,95,100,105,110,115,120,125],{"id":81,"slug":82,"title":83,"created_at":84},"d4cffde7-9b50-4cc7-bb68-8bc9e3b15477","nvidia-rubin-ai-supercomputer-en","NVIDIA Unveils Rubin: A Leap in AI Supercomputing","2026-03-25T16:24:35.155565+00:00",{"id":86,"slug":87,"title":88,"created_at":89},"eab919b9-fbac-4048-89fc-afad6749ccef","google-gemini-ai-innovations-2026-en","Google's AI Leap with Gemini Innovations in 2026","2026-03-25T16:27:18.841838+00:00",{"id":91,"slug":92,"title":93,"created_at":94},"5f5cfc67-3384-4816-a8f6-19e44d90113d","gap-google-gemini-ai-checkout-en","Gap Teams Up with Google Gemini for AI-Driven Checkout","2026-03-25T16:27:46.483272+00:00",{"id":96,"slug":97,"title":98,"created_at":99},"f6d04567-47f6-49ec-804c-52e61ab91225","ai-model-release-wave-march-2026-en","Navigating the AI Model Release Wave of March 2026","2026-03-25T16:28:45.409716+00:00",{"id":101,"slug":102,"title":103,"created_at":104},"895c150c-569e-4fdf-939d-dade785c990e","small-language-models-transform-ai-en","Small Language Models: Llama 3.2 and Phi-3 Transform AI","2026-03-25T16:30:26.688313+00:00",{"id":106,"slug":107,"title":108,"created_at":109},"38eb1d26-d961-4fd3-ae12-9c4089680f5f","midjourney-v8-alpha-features-pricing-en","Midjourney V8 Alpha: A Deep Dive into Its Features and Pricing","2026-03-26T01:25:36.387587+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"bf36bb9e-3444-4fb8-ab19-0df6bc9d8271","rag-2026-indispensable-ai-bridge-en","RAG in 2026: The Indispensable AI Bridge","2026-03-26T01:28:34.472046+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"60881d6d-2310-44ef-b1fb-7f98e9dd2f0e","xiaomi-mimo-trio-agents-robots-voice-en","Xiaomi’s MiMo trio targets agents, robots, and voice","2026-03-28T03:05:08.899895+00:00",{"id":121,"slug":122,"title":123,"created_at":124},"f063d8d1-41d1-4de4-8ebc-6c40511b9369","xiaomi-mimo-v2-pro-1t-moe-agents-en","Xiaomi MiMo-V2-Pro: 1T MoE Model for Agents","2026-03-28T03:06:19.238032+00:00",{"id":126,"slug":127,"title":128,"created_at":129},"a1379e9a-6785-4ff5-9b0a-8cff55f8264f","cursor-composer-2-started-from-kimi-en","Cursor’s Composer 2 started from Kimi","2026-03-28T03:11:59.132398+00:00"]