Apple Launches Siri AI in iOS 27 Powered by Gemini-Refined Foundation Models

Apple shipped iOS 27 on September 14, 2026, delivering the first public release of Siri AI — a completely rebuilt assistant that replaces a decade of intent-matching with a five-model generative AI architecture trained partly on outputs from Google's Gemini frontier models. iOS 27 is available for every iPhone from the iPhone 11 onward, but Siri AI itself requires at minimum an iPhone 15 Pro or iPhone 16, launching in English-only beta with undisclosed daily usage caps, no access on iPhone or iPad in the EU, and no access anywhere in China. For the hundreds of millions of users on qualifying hardware, the old keyword-and-shortcut Siri is gone.

How Apple Rebuilt Siri From the Ground Up

Siri AI is powered by the third generation of Apple Foundation Models — a family of five purpose-built models that Apple co-developed with Google following a partnership announced in January 2026. According to Apple's AI VP Amar Subramanya, every model in the AFM 3 family was trained using Apple's proprietary data through reinforcement learning, then refined using outputs from Gemini frontier models — a distillation process in which a larger model's knowledge is used to guide training of a smaller one. Apple SVP Craig Federighi clarified the boundaries of that collaboration explicitly: "The amount of the Google Assistant we use is none," he said at a post-keynote press briefing, explaining that Apple uses none of Google's client-side code, none of the Gemini models deployed to Google's consumers, and no Google Search infrastructure as the knowledge backbone for Siri.

The five models divide across two deployment environments. On-device, AFM Core is a next-generation dense model in roughly the 3-billion-parameter range, designed as the workhorse for everyday tasks on qualifying Apple Intelligence devices — iPhone 15 Pro, iPhone 16, or anything newer. AFM Core Advanced is a higher-power on-device model that enables expressive voice synthesis and significantly improved dictation accuracy without any network round trip; it requires higher-end hardware: iPhone 17 Pro, iPhone Air, iPhone 18 Pro or Pro Max, or a Mac with an M3 chip and at least 12GB of unified memory.

On the server side, three models run through Apple's Private Cloud Compute infrastructure, where — according to Apple — user requests are processed without data being stored or made accessible to Apple or Google. AFM Cloud handles latency-sensitive cloud requests, AFM Cloud Image powers the new photorealistic generation in Image Playground and the AI photo editing in Photos, and AFM Cloud Pro — the most capable model in the family — is designed specifically for multi-step agentic tasks and complex reasoning, with quality Apple describes as comparable to Gemini frontier models. AFM Cloud Pro runs on Google's cloud TPUs combined with NVIDIA GPUs, all operating under Private Cloud Compute certification that independent security researchers can verify.

The system orchestrator that ties these models together operates entirely on-device. It coordinates requests between Siri AI and two on-device indexes: the Spotlight index, which tracks content across Mail, Messages, Photos, Notes, and other apps, and the App Toolbox, which encodes what actions Siri can take inside each application. This on-device coordination is the architectural feature that allows Siri to retrieve a family member's email from six months ago, add its ingredients to a Reminders list, and confirm the calendar invite — all in one conversation — without sending the content of that email to Apple's servers for routing decisions.

What Siri AI Can Do Across the Apple Ecosystem

In practice, Siri AI operates in three modes that did not exist in the previous version of Siri. First, it draws on personal context: the system can search across a user's messages, emails, photos, and calendar to find relevant information in response to natural-language questions. Second, it operates with onscreen awareness on every supported Apple platform — on iPhone via a new Siri mode in the Camera app that allows real-world object queries, on iPad integrated directly into the screenshot experience, on Mac through a dedicated keyboard shortcut that lets users type queries about anything visible on their display, and on Apple Vision Pro through gaze-based invocation that responds to what a user is physically looking at. Third, it supports conversational follow-up: users can extend any Siri response into a multi-turn dialogue, and a new dedicated Siri app stores conversation history — synced end-to-end encrypted via iCloud — so a thread started on iPhone can continue on Mac, iPad, or Apple Watch.

The rebuilt Writing Tools integrated into Siri AI function across virtually every text field in the system, including most third-party apps. A user can describe what they want written, and Siri generates a draft; they can describe changes, and Siri rewrites. In Mail and Messages, Siri draws on the user's prior communication patterns to match the tone and punctuation style they typically use with each recipient — a personalization layer that distinguishes it from a generic AI writing tool.

For third-party developers, Siri AI launched with existing integrations intact — WhatsApp, Audible, and others remain supported through the App Toolbox interface. Apple announced that additional integrations with Microsoft Outlook, Notability, and Tripsy are in development, which will allow Siri to draft Outlook emails, search for documents in Notability, and add restaurant recommendations to Tripsy without the user switching apps.

Siri AI Arrives Two Years Late to a Market Gemini Already Leads

Apple first promised the personal-context Siri overhaul at WWDC 2024 — alongside the iPhone 16's launch that September. The features were not ready. Through 2025 and into early 2026, Apple repeatedly delayed the more capable Siri, prompting criticism from users who had upgraded their hardware partly on the promise of the AI features. The Siri AI available today is what Apple committed to delivering in the fall of 2024.

In the intervening two years, Google Gemini became the default assistant on Android and replaced Google Assistant entirely. Gemini Advanced ($19.99 per month with Google One AI Premium) has offered multi-turn conversational AI, real-time voice interaction via Gemini Live, and multimodal image understanding as consumer products since 2024. Samsung's Galaxy AI, built on on-device Gemini Nano plus cloud Gemini, ships on Galaxy S24 and S25 hardware and includes features like Circle to Search and real-time call translation. OpenAI's ChatGPT for iOS features voice mode with near-real-time conversation.

The architectural distinction Apple emphasizes — OS-level integration that reads personal data by design, versus a separately launched app — is genuine. No competing assistant operates at the same depth within the iOS permission model: Siri AI can read messages, search emails, and execute cross-app actions through the App Toolbox without requiring the user to explicitly authorize each data lookup. That integration advantage is Apple's strongest technical argument for Siri AI against competitors who, however capable, remain separate applications the user must consciously invoke.

The Features Apple Has Not Disclosed: Usage Caps and Paid Tiers

Apple's iOS 27 announcements include language that carries more commercial significance than any individual Siri AI capability: according to Apple's official product page, certain Apple Intelligence features — including image generation and features that rely on powerful server models — are subject to daily usage limits. Expanded access is available with most iCloud+ subscription plans. Apple has confirmed that higher iCloud+ tiers will include expanded Apple Intelligence access; current iCloud+ plans range from $0.99 to $9.99 per month.

Apple has not disclosed the numerical ceiling of those limits — not in any product page, not in any developer documentation available at launch. It has also not disclosed the price of the future paid tier, the date that tier will become available, or how users will be notified when they approach or hit the daily cap.

The significance of this structure extends beyond pricing mechanics. It marks the first time Apple has built a core iOS feature — the main assistant — around a consumption model that will eventually require a subscription to use without limits. Apple has previously offered services as subscriptions (Apple Music, iCloud storage, Apple TV+), but the operating system's primary interface layer has always been free. Siri AI begins to change that boundary. Users who invest heavily in Siri AI for daily workflows — drafting emails, retrieving context, triggering complex automations — will eventually encounter an upsell moment within the same UI where they ask Siri to add milk to a grocery list.

EU and China Blocks Leave Hundreds of Millions Without the Feature

Siri AI is not available on iPhone, iPad, or Apple Watch in the EU at launch. EU Mac and Apple Vision Pro users can access it when their device language is set to English. Apple states it is "working hard to find a path forward that preserves its users' privacy and security" — language that frames the EU absence as a privacy position rather than a regulatory compliance shortfall, though the underlying cause is unresolved Digital Markets Act (DMA) obligations Apple has not met to regulators' satisfaction. No launch date for EU iPhone availability has been given.

In China, Siri AI is absent entirely. Apple cited regulatory requirements without specifying which ones or providing a timeline. The combination means that two of the world's largest smartphone markets — the EU representing over 150 million iPhones in active use, China representing Apple's largest single market by unit volume — have no access to the headline AI feature of Apple's 2026 software platform.

Audio Intelligence, Health AI, and the Features Still Coming Later This Year

Several capabilities announced under the Apple Intelligence umbrella are not in today's release. Audio Intelligence — including Live Rewind (which shows the last 15 seconds of a nearby conversation as text when a user double-presses the Apple Watch Digital Crown) and Siri Recap (which generates a brief summary of recent conversations for review) — requires Apple Watch Series 12 or Ultra 4 and will arrive in beta later in 2026. Both rely on Secure Exclave, a hardware-isolated compartment inside the Watch's S11 chip that processes audio and immediately deletes it, keeping the raw audio inaccessible to the OS, apps, or Apple.

The redesigned Health app, which uses Apple Intelligence to turn longitudinal health data into personalized guidance — surfacing an Insights tab that updates throughout the day across heart, sleep, fitness, and cycle tracking data — is similarly slated for a later 2026 update.

The SynthID standard, which would allow users to identify images generated or edited by Apple Intelligence tools, will arrive in a software update later this year rather than today.

Today's release of Siri AI is therefore the beginning of the Apple Intelligence 2026 rollout, not its conclusion. Apple has built a credible architectural foundation — five models, Private Cloud Compute certification, deep OS integration, end-to-end encrypted conversation memory — but it has shipped the centerpiece feature as a beta with undisclosed limits, in one language, excluded from two major markets, and with its most technically novel capabilities (Audio Intelligence, the full Health AI redesign) deferred to later in the year. How quickly Apple resolves the EU regulatory impasse, how aggressively it prices the eventual paid tier, and whether independent evaluators confirm the company's capability claims without published benchmarks will determine whether today marks the inflection point Apple is presenting it as — or simply the start of an unusually long beta period.