After a prolonged wait, Siri’s AI upgrade has finally made its debut.

In a 75-minute keynote, with 70% of the focus dedicated to AI, Cook appeared somewhat ‘eager’ during his final WWDC as host.

WWDC stands as one of Apple’s most pivotal annual events, unveiling the latest advancements in software systems such as iOS, iPadOS, and macOS. Over the past three years, AI has emerged as the central theme at WWDC, shaping the future of Apple’s operating systems.

Two years ago, Apple boldly introduced Apple Intelligence, aiming to redefine intelligent assistants amid the AI surge. However, progress has been sluggish, with the new Siri experiencing repeated delays.

This year’s WWDC 2026 carries the weight of restoring Apple’s AI credibility. From Cook’s opening remarks, it’s clear: the long-awaited Siri is finally here.

The highlight of this year’s WWDC is undoubtedly the arrival of the much-delayed Siri.

Apple has rebranded the upgraded Siri as Siri AI. No longer just a ‘voice remote’ for playing songs, making calls, or checking the weather, it has evolved into a true system-integrated Agent.

In essence, the new Siri AI is no longer underperforming. Like last year’s revolutionary Doubao smartphone, Apple’s mobile AI now boasts system-wide capabilities.

The new Siri boasts five core capabilities: personal context understanding, image comprehension, world knowledge, screen awareness, and APP invocation. Nearly all daily tasks can be accomplished through Siri, from replying to emails and scheduling appointments to drafting articles based on chat histories.

In a demo, the presenter opened a photo, and Siri instantly recognized the location on the screen, navigated to it on the map, retrieved an address shared by a friend on a social app, and planned a route with a stopover at the friend’s place—all in one seamless flow. Clearly, the new Siri has become a higher-level gateway above apps.

Another demonstrator used Siri to check the World Cup’s first-week schedule, proposed hosting a Brazil vs. Morocco viewing party, and asked Siri to recommend classic dishes from both countries. Siri then searched global knowledge, retrieved coconut cookies mentioned by friend Maria in a chat, and compiled a menu blending both cuisines. It even drafted a group invitation with the menu and sent it with a single tap—all without manual intervention.

According to Apple’s official demos, Siri AI is no longer a mere assistant. With Agent capabilities and multimodal functions, it can now ‘see’ the phone screen and ‘act’ on behalf of users. While these tasks are primarily based on Apple’s native apps and haven’t yet reached the level of coordinating third-party apps—or even matching the freedom of domestic AI phones to order coffee with a single phrase—this is still a significant leap for Apple, which has ‘admitted to being slow on AI.’

Additionally, Apple showcased a series of AI-powered ‘standalone’ app capabilities, underscoring AI’s versatility.

Siri can also assist in organizing photo albums. A simple command like ‘Put photos with X in the family-shared album’ allows Siri to complete identification, filtering, and operations without opening the app.

Siri is also integrated into the camera, enabling users to ask questions by pointing the lens at an object, such as inquiring about the calorie count of a dish or splitting a bill among friends.

Regarding activation methods and multi-device experiences, Siri AI retains both ‘Hey Siri’ and side-button wake-up options while being embedded in the Dynamic Island. A downward swipe enables multi-round voice or text conversations. Furthermore, Apple has launched a dedicated Siri app, with all conversation records privately synced via iCloud. Dialogues initiated on an iPhone can continue on an iPad and conclude on a Mac.

Siri’s form varies across platforms. On Mac, Siri is integrated into Spotlight, accessible from any interface, and allows users to ask questions about selected content via right-click menus. On Apple Watch, Siri AI executes operations directly with minimal interaction. Vision Pro takes it further, requiring only a glance at Siri to speak—no wake-up phrase needed.

These capabilities are powered by Apple Intelligence. The system operates on-device and in private clouds, handling simple tasks locally and complex ones via Private Cloud Compute.

Built on the Apple Intelligence foundation, native apps like Safari, Messages, Mail, and Calendar are now infused with AI capabilities.

Behind Siri AI’s impressive performance lies Apple’s complete rebuild of its AI architecture, finally delivering the two-year-delayed Apple Intelligence.

Recall that at WWDC 2024, Apple’s high-profile launch of Apple Intelligence raised expectations. Promising the most ‘Apple-like’ experience, seamless AI large model collaboration (ChatGPT), and smooth end-to-end cloud integration, Apple set a benchmark for AI smartphone vendors. Some analysts and consumers even believed Apple might still develop its own AI.

However, at WWDC 2026, all doubts were dispelled. Apple Intelligence is built on Google Gemini.

While details remain limited, we can still examine Apple’s new AI architecture. Logically, Apple AI adopts the industry-consensus end-to-end cloud-collaborative framework. Apple’s foundational models—Apple Foundation Models—are a series co-developed with Google based on Gemini, deployed on-device and in the cloud (Private Cloud Compute).

On the cloud side, Apple has built dedicated AI infrastructure. The company claims the cloud only processes user requests and ‘deletes’ data afterward, ensuring Apple neither stores nor accesses user information.

This design is unconventional, as AI large model vendors typically value user interaction data to iteratively improve models. For Apple, however, it seems like a ‘one-and-done’ approach for complex user demands (image generation, complex reasoning, etc.). This may suggest Apple lacks a mature data loop or even independent model training capabilities.

On the device side, Apple categorizes models into high- and low-capability tiers this year. All Apple Intelligence-compatible devices come with a ~3B-parameter foundational model.

On higher-performance devices (e.g., latest phones, PCs), Apple adds a larger model capable of higher-quality outputs and longer contexts. It also includes a dedicated speech model for natural conversations and personalized voice synthesis in the new Siri.

Apple’s solution for on-device foundational models is noteworthy. It introduced a System Orchestrator architecture to manage Apple Intelligence.

To clarify, an Orchestrator in AI (especially in the Agent era) coordinates tasks across multiple small models with different capabilities, tools (search, APP invocation), end-to-end cloud tasks, and contextual (multi-step) memory. The Orchestrator’s role is to break down overall tasks into suitable sizes and assign them to the most appropriate components.

Apple’s System Orchestrator manages four functional modules: personal information understanding, world knowledge, Actions, and screen awareness.

Specifically, contextual understanding refers to on-device information—text, images, emails—which the mobile AI comprehends fully to invoke the right data.

World knowledge represents general common sense. Apple offers an online world knowledge service to supplement local model gaps or provide up-to-date information when needed.

Actions enable Siri AI to perform operations beyond chatting, acting as users’ ‘hands’ to manipulate phones.

Screen awareness serves as users’ ‘eyes,’ reading on-screen information for AI model input.

Overall, the four modules form two paired structures: internal/external information acquisition and execution output/information input. According to Apple, this is AI centered around you.

After two years, Apple has finally delivered Apple Intelligence. However, its AI still feels somewhat ‘conflicted,’ from application effects to strategic layout.

On one hand, Apple Intelligence retains many ‘legacy’ AI assistant features. For example, the world knowledge component is a knowledge graph Apple has operated for years, initially aimed at addressing Siri’s outdated or fabricated information. In today’s era of connected AI large models, this seems redundant. On the other hand, Apple’s decision to partner with Google on AI models marks a departure from its obsession with in-house development.

Nevertheless, any AI implementation is a positive step. Especially in 2026, as Chinese smartphone vendors showcase their AI prowess, Apple has finally grasped the importance of delivery over perfection.

Finally, the update roadmap: Apple AI supports devices as old as the iPhone 11, but PCs must use Apple chips. China and the EU are excluded for now.

However, Chinese users may gain access soon. Previously, Apple planned to partner with Baidu, using ERNIE Bot 4.0 as the foundation for generative AI on Chinese iPhones to comply with domestic data regulations. It later collaborated with Alibaba for AI compliance reviews in China.

Without direct access to Google services, Apple will likely seek a local AI partner, with QianWen being the most probable candidate.

The success and speed of this partnership will directly determine whether Chinese users get the new Siri.

Beyond regional restrictions, what truly matters at this WWDC is Apple’s AI-era reconstruction and the integration of Agents into its operating system.