iOS 27 AI: 12GB Requirement, Siri Rebuild, and What Your iPhone Gets

At WWDC 2026, Apple revealed iOS 27 with its most ambitious AI overhaul yet. But the update comes with a hardware divide that leaves many recent iPhones on the sidelines for the best features. The new on-device AI model, a redesigned Siri called Siri AI, and a framework that welcomes third-party chatbots mark a fundamental shift in Apple's AI strategy. Here is what was actually announced and which devices make the cut.

The 12GB On-Device AI Requirement

Apple raised the bar for on-device AI performance. The company's most powerful on-device model in iOS 27 requires a minimum of 12GB of unified memory, double the 8GB that was previously sufficient for Apple Intelligence .

This threshold creates a clear split in the iPhone lineup:

• Supported: iPhone Air, iPhone 17 Pro, and iPhone 17 Pro Max are the only iPhones that can run the most advanced model locally .
• Excluded: The standard iPhone 17, despite being released the same generation, ships with 8GB of memory and cannot run the model . This also leaves out iPhone 16 Pro, iPhone 15 Pro, and all older devices.

The limitation affects several headlining features. Craig Federighi, Apple's SVP of Software Engineering, stated during the keynote that the most powerful on-device model and the capabilities it drives—specifically expressive customizable voices and a significant jump in dictation accuracy—require Apple's most capable hardware .

The new on-device model itself is a technical leap forward. It gains vision capabilities, accepting image input for the first time, and doubles its context window to 8,192 tokens on newer devices, up from 4,096 in iOS 26 . It also features improved instruction following and better tool calling .

For broader compatibility, the full list of devices that support the 12GB model includes:

• iPad: Models with M4 and later with 12GB or more of memory .
• Mac: Models with M3 and later with 12GB or more of memory .
• Apple Vision Pro: The model with the M5 chip .

Devices that fall short can still access most of the same Apple Intelligence features, but they will rely on Apple's Private Cloud Compute (PCC) servers instead of running them locally, which may result in slower responses .

Siri AI: Built on Google Gemini

The most significant AI change in iOS 27 is Siri AI, a ground-up rebuild of the voice assistant. Apple's internal models were reportedly not ready for the scale of the overhaul, leading the company to license Google's Gemini models to power the core reasoning engine through a deal reported to be worth up to $1 billion per year .

Siri AI introduces several major new capabilities:

• Standalone App: For the first time, Siri has its own dedicated app with a chat-style interface that includes text input, a microphone toggle, and a paperclip for attachments. Conversation history syncs across your devices via iCloud .
• On-Screen and Personal Context: The assistant can now reason about what is visible on your screen and pull answers from your personal data, including emails, messages, calendar, notes, and photos .
• Persistent Interface: Siri's visual presence moves to the Dynamic Island as an always-on agent. A new swipe-down "Search or Ask" gesture provides system-wide activation .

A Platform for Rivals: The Extensions Framework

In a historic shift, iOS 27 introduces an Extensions framework that allows users to set a third-party AI chatbot as the default engine powering Siri, Writing Tools, and Image Playground .

This means that instead of being locked into Apple's own AI, users can choose from options like:

• Google Gemini
• Anthropic Claude
• OpenAI ChatGPT
• Custom enterprise agents

Third-party AI providers can plug in through the Extensions system and, with user permission, gain access to deep system integrations such as on-screen awareness and personal context—the same capabilities Apple's own assistant has . The system is designed so that pressing the side button can trigger a conversation with the AI service of your choice .

Daily Usage Caps and the Three-Tier Privacy Stack

The new AI features are not entirely unrestricted. Apple imposes daily usage caps on server-side AI features, including image generation. Users who subscribe to a paid iCloud+ plan, however, will receive increased limits .

To handle user data, Apple described a three-tier privacy architecture at WWDC :

1. On-Device: The simplest tasks are processed directly on your iPhone or iPad using Apple's own models. There are no request limits for on-device operations .
2. Private Cloud Compute (PCC): Moderately complex requests that require more power are routed to Apple's secure server infrastructure, which offers a 32,768-token context window . Data is processed statelessly, meaning it is not retained or stored .
3. Google Cloud: The heaviest reasoning tasks are routed to Google's cloud servers. Apple stated that Google is contractually barred from using any of this data for training or any other purpose .

Launch Timeline

All of these features are scheduled to arrive as part of the public release of iOS 27 in September 2026 .

Siri AI itself will initially launch as a beta in US English later in the year before the full public rollout, with support for more languages to follow . It is important to note that Siri AI will not be available in the European Union at launch due to the Digital Markets Act, nor will it launch in China initially .

The bottom line is a segmented release: while iOS 27 installs on devices as far back as the iPhone 11, the transformative "on-device" AI experience is an exclusive club for the latest high-end models.