Apple’s September 9 fall event put AI ahead of hardware. Across the keynote and six official press releases, the company rolled out new artificial intelligence features for Siri, Apple Watch, the Health app, and the camera. Two details stand out early for users: Siri AI will arrive in English first, and cloud-based Apple Intelligence features will carry daily usage limits, with higher quotas set to become paid later on.
Siri now handles more than 2.5 billion requests a day
Apple said during the event that Siri processes more than 2.5 billion requests every day. The new Siri AI is built on Apple’s latest foundation model, and the company said it can already work smoothly with more than 300,000 apps.
Apple grouped the system’s capabilities into three broad layers. The first is personal context. In the onstage example, a user asked what their mother had mentioned wanting to do during a visit, and Siri searched across Mail and messages to find the answer. The second is action-taking. Apple showed third-party examples that included sending a message in WhatsApp, playing an audiobook in Audible, drafting an email in Outlook, finding homework in Notability, and adding a restaurant recommendation to Tripsy. The third layer is broader world knowledge through a connection to the web.
Apple also introduced a standalone Siri app placed directly on the home screen. Conversations are collected there and synced privately through iCloud, so a user can start on an iPhone and continue on an iPad. Threads can also be pinned. Siri is now built into the Camera app as well, with a new camera mode that lets Siri answer questions about what the user is looking at.
Siri AI is separate from broader Apple Intelligence language support
One of the clearest caveats concerns language availability. Traditional Chinese is included in the 16 languages Apple Intelligence will support with iOS 27, but Siri AI is on a different track. Apple’s Taiwan website says plainly that Siri AI will debut in English first and that usage may be limited.
That means Siri AI will launch as a beta alongside iOS 27, but only in English at first. Apple’s new watch audio intelligence features and the redesigned Health app are also scheduled for later this year and will start in English as well. Some of the most eye-catching AI features from the event will not be available in a Traditional Chinese environment in the near term.
Apple says server-side AI has daily caps and higher usage will be paid
A second point appeared in a press release footnote. Apple said Apple Intelligence features that rely on server-side models come with daily usage limits. The list includes Siri AI, intelligent photo editing tools, Image Playground, and cloud models used in Shortcuts. The cap can vary depending on the feature, request complexity, system load, and system policy.
Apple then added the key line: higher usage limits for these features will be offered in the future through a paid option. It is the first time Apple has explicitly said Apple Intelligence will have a paid tier.
The company did not disclose pricing, rollout timing, or the size of the limits. On-device models are not affected. Apple said tasks that can run locally will stay on the device, while jobs that need more compute will be sent to Private Cloud Compute.
Apple Watch adds audio intelligence
Series 12 and Ultra 4 use the S11 chip to introduce a new class of Apple Intelligence features that Apple calls audio intelligence. The features are set to arrive in beta later this year. Apple listed four of them.
- Sound Recognition: on-device detection of important sounds such as sirens, alarms, doorbells, and babies crying, with alerts sent to the user even when the iPhone is not nearby. Apple said this feature is designed for deaf and hard-of-hearing users.
- Live Rewind: a double-click of the Digital Crown shows text snippets from the previous 15 seconds of conversation so users can catch something they missed or did not understand. Those snippets can be queried with Siri again or saved into the Siri app.
- Siri Recap: when turned on, Apple Intelligence automatically creates a title and key points in the Siri app after a conversation ends.
- Shazam: identifies music playing nearby and shows the song title and artist directly in the music recognition widget in Smart Stack.
Secure Exclave isolates audio processing
Because these features listen to environmental sound and conversation, Apple spent unusual time on privacy architecture. The S11 chip includes Secure Exclave, a dedicated isolated hardware block for audio processing. Audio is handled in complete separation from the rest of the system and deleted immediately after processing.
Apple said the features do not create or store recordings. The raw audio used for processing is not accessible to the operating system, apps, the user, or Apple itself.
Apple also built in deliberate limits. Audio intelligence does not identify or assign speakers, which the company said is meant to protect both the user and nearby people. Siri Recap produces a short high-level summary rather than a transcript, and it is designed to exclude sensitive material such as financial information and government-issued identification numbers.
There is also a notice for third parties. When Live Rewind is activated, the watch emits an alert sound even if it is muted. It also shows a full-screen animation and microphone indicator so nearby people can hear and see that the feature is running. All of these functions are opt-in. Siri Recap can be switched on or off from Control Center. Live Rewind text snippets and Siri Recap summaries sync through iCloud with end-to-end encryption inside the Siri app, and Apple said it cannot access them.
The Health app will generate AI insights
The Health app on iPhone is due for a redesign later this year. A new Insights tab will update summaries throughout the day, covering heart data, sleep, readiness, fitness, vital signs, and cycle tracking.
A separate Recommended for You section will use personal data to offer suggestions and related articles. Apple’s example was a recommendation to adjust a bedtime wind-down routine after several late nights in a row.
The new Longevity tab includes a Health Age feature that compares a user’s measured indicators with their actual age. These features sit on top of a new health sensing system in Series 12 and Ultra 4. Apple said the watches use a larger and more power-efficient green LED, allowing all-day heart-rate readings every five seconds. Heart-rate variability can be measured as often as every five minutes, which Apple said is 24 times more frequent than before.
Apple also cited a study involving more than 1,000 participants, saying Apple Watch had the highest heart-rate sensing accuracy among the tested devices.
Camera AI ranges from auto-shutter to tamper-resistant reference images
On the camera side, Smart Take on iPhone Duo uses an on-device AI model on A20 Pro to analyze the scene in real time and trigger the shutter automatically once the subject is ready. Apple presented it as a way to solve the problem of the photographer being left out of the shot or rushing to beat a self-timer.
For iPhone 18 Pro, Apple introduced Apple Reference Image. The new main camera sensor signs image data pixel by pixel at the moment a photo is taken. In Reference mode, that signed data is processed through Private Cloud Compute into a reference image that cannot be altered. It is stored in the Photos app alongside the normal editable image and works like a digital negative.
Apple also said it will support the SynthID standard in the future to identify images generated or edited by AI.
Taken together, the two features show Apple moving in both directions at once: using generative AI to help users capture photos, and building a mechanism that lets a photo prove it has not been changed. Apple’s press release said the latter is intended for photojournalists and photographers.
A20 Pro powers the system with dual 16-core neural engines
Much of the compute behind these features comes from A20 Pro. Apple said the chip uses a 2-nanometer process and includes a new dual 16-core neural engine setup, for 32 cores in total. The company said its AI processing capability is twice that of A19 Pro.
The 6-core CPU now integrates a neural accelerator directly into the cores, and memory bandwidth is 50% higher than A19 Pro. Apple said iPhone 18 Pro can access its most advanced on-device foundation model, while tasks beyond device capacity are sent to Private Cloud Compute. The on-device model also improves system-level dictation accuracy.
Developers can use Apple’s on-device models at no cost
For developers, one of the central announcements is the Foundation Models framework. According to Apple Intelligence’s official page, any app can use the on-device model that powers Apple Intelligence. Features built with it can work offline and are not charged per request.
That creates a sharp split with the cloud usage model Apple described earlier: inference on a user’s device is free, while inference on Apple’s servers is where usage caps and future charges apply. Developers can also use App Intents and related APIs to integrate Siri, Writing Tools, and Image Playground into their own apps.
Release schedule
Apple said iOS 27 will roll out as a free update on September 14, Pacific Time. Apple Intelligence will go live the same day with support for 16 languages. Siri AI will also launch with iOS 27 in beta, but in English first. Apple Watch audio intelligence and the redesigned Health app are both scheduled for later this year, also starting in English.