跳到主要內容

Siri Gemini評測:蘋果砸10億美元外包AI,賭硬體翻身 | Siri Gemini Review: Apple Pays $1B/Year, Bets on Hardware

By Kit 小克 | AI Tool Observer | 2026-09-07

🇹🇼 Siri Gemini評測:蘋果砸10億美元外包AI,賭硬體翻身

蘋果新執行長 John Ternus 9月1日正式上任,第一件事就是把最尷尬的問題攤在陽光下:Siri追不上對手,乾脆花錢跟Google買技術。Siri Gemini是蘋果與Google去年底敲定、今年初正式公布的合作案,讓Google客製化的Gemini模型接手重建Siri的大腦,傳出每年要付Google約10億美元授權費。這篇文章帶你看懂Siri Gemini到底是什麼、怎麼運作,以及蘋果為什麼寧願外包AI核心,也要保住硬體霸主的位子。

什麼是Siri Gemini?蘋果為何找Google合作?

Siri Gemini是蘋果把Google客製化的1.2兆參數Gemini模型,接進重建版Siri的技術架構,讓Siri具備螢幕理解、跨App推理與對話記憶等能力。蘋果自研模型多年做不出打得贏對手的成果,乾脆認清現實找外援。

根據報導,新版Siri複雜多步驟任務的完成率從58%拉高到92%,回應時間壓在0.5秒內,內部代號「Project Campos」。第一批Gemini驅動功能已隨iOS 26.4上線,更完整的對話式Siri則隨iOS 27在9月正式推出。

Siri Gemini安全嗎?資料會不會被Google拿走?

蘋果強調,Gemini查詢會透過自家Private Cloud Compute處理,採端對端加密與硬體隔離環境運算,模型權重跑在蘋果基礎架構裡,Google拿不到使用者資料——這是蘋果談這筆交易時最在意、也最常拿出來說服使用者的一點。

Ternus的AI策略:不拚模型,拚硬體

比起燒錢死磕前沿模型,Ternus的賭注是「感測器+裝置」比雲端大模型更懂使用者——例如AirPods上的鏡頭,比純文字聊天機器人知道更多你當下在做什麼。蘋果接下來的硬體布局包括:

  • 內建鏡頭的AirPods,即時讀取環境畫面
  • 智慧眼鏡,正面對標Meta的Ray-Ban AI眼鏡路線
  • 家用機器人顯示裝置,把Siri Gemini搬進客廳
  • 9月9日發表的可折疊iPhone Ultra,是Ternus上任後第一場硬體大考

小克怎麼看Siri Gemini?

老實說,蘋果肯認錯、外包AI大腦,比死撐自研模型健康。Siri Gemini解決的是「腦袋不夠聰明」的問題,但蘋果真正的籌碼從來不是模型參數量,而是十幾億支裝置的分發能力。這場賭局能不能贏,9月9日的新機發表會就是第一個檢查點。

好不好用,試了才知道。


🇺🇸 Siri Gemini Review: Apple Pays $1B/Year, Bets on Hardware

Siri Gemini is what everyone's calling Apple's new AI stack: a custom 1.2-trillion-parameter Google Gemini model quietly powering a rebuilt Siri, under a deal reportedly worth around $1 billion a year. It's also the first real test for Apple's new CEO John Ternus, who took over on September 1 and immediately inherited the one problem Apple couldn't out-engineer on its own — a frontier AI model.

What Is Siri Gemini and Why Did Apple Partner With Google?

Siri Gemini plugs a customized Gemini model into Siri's backend, adding screen understanding, cross-app reasoning, and conversational memory that Apple's in-house models couldn't deliver after years of trying.

Reported benchmarks show complex multi-step task completion jumping from 58% to 92%, with sub-0.5-second responses. Internally codenamed "Project Campos," the first Gemini-powered features shipped with iOS 26.4, and the fuller chatbot-style Siri rolled out with iOS 27 this September.

Is Siri Gemini Private? Does Google See Your Data?

Apple says Gemini queries run through its own Private Cloud Compute — end-to-end encrypted, hardware-isolated enclaves — with model weights hosted on Apple's own infrastructure, not Google's. No user data reportedly leaves Apple's control, which is the pitch Apple leans on hardest.

Ternus's Real Bet: Hardware Over Frontier Models

Ternus's thesis is that sensor-rich devices beat a cloud chatbot at knowing the user — a camera on your AirPods sees more context than any prompt ever could. That bet shows up across Apple's roadmap:

  • Camera-equipped AirPods for real-time environmental awareness
  • Smart glasses competing directly with Meta's Ray-Ban line
  • A robotic home display bringing Siri Gemini into the living room
  • The foldable iPhone Ultra launching September 9 — Ternus's first big hardware test

Kit's Honest Take on Siri Gemini

Admitting Siri needed someone else's brain is healthier than pretending Apple's in-house model was fine. Siri Gemini fixes the "not smart enough" problem, but Apple's real leverage was never model size — it's over a billion devices to distribute through. Whether this bet pays off starts getting tested at the September 9 event.

好不好用,試了才知道 — good or not, you won't know until you try it.

Sources / 資料來源

常見問題 FAQ

Siri Gemini是什麼?

Siri Gemini是蘋果把Google客製化的1.2兆參數Gemini模型接入重建版Siri,讓Siri具備螢幕理解、跨App推理與對話記憶,用來彌補蘋果自研AI模型落後的問題。

Siri Gemini什麼時候上線?

第一批Gemini驅動功能隨iOS 26.4上線,更完整的對話式Siri則隨iOS 27在2026年9月正式推出。

Siri Gemini會把使用者資料傳給Google嗎?

蘋果表示不會。Gemini查詢透過蘋果自家Private Cloud Compute處理,採端對端加密與硬體隔離環境,模型權重跑在蘋果基礎架構上,Google無法取得使用者資料。

蘋果每年要付Google多少錢?

根據Bloomberg等媒體報導,這筆授權合作約每年10億美元,讓蘋果取得客製化的1.2兆參數Gemini模型使用權。

新執行長Ternus的AI策略是什麼?

Ternus押注硬體而非前沿模型:透過內建鏡頭的AirPods、智慧眼鏡、家用機器人顯示裝置等感測器豐富的裝置,讓蘋果比純雲端聊天機器人更懂使用者的即時情境。

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code