← Back to feed

Apple will debut a completely rebuilt Siri at WWDC 2026 on June 8, according to a May 12 Bloomberg report from Mark Gurman backed by 9to5Mac, MacRumors, Gizmodo, and Cult of Mac. The overhaul scraps the hybrid voice-assistant architecture that Craig Federighi acknowledged had failed internally in August 2025 and replaces it with a system-wide AI agent running on Google Gemini under a multi-year $1 billion annual licensing deal signed January 2026.

The new Siri is designed as an “always-on agent that can tap into personal data and take action across apps” — language that positions it directly against Claude, ChatGPT, and Gemini’s consumer products for the first time since Siri launched in 2011.

What’s Changing

Standalone app. For the first time in 15 years, Siri gets its own dedicated application, codenamed “Rave.” It surfaces conversation history in a grid of rounded-rectangle previews, supports pinned chats, file and image uploads, and a chatbot-style thread view matching iMessage’s visual language. Users interact via text or voice and can switch modes mid-conversation.

Dynamic Island as entry point. Activating Siri shows a “Search or Ask” prompt in the Dynamic Island. A swipe down from the top-centre of any screen opens the same interface system-wide, replacing Spotlight Search with a richer inference layer that pulls from installed apps and the open web.

Cross-app agent actions. The new Siri executes multi-step tasks across Apple’s first-party apps and, via a new extensions framework, in third-party apps. An “Ask Siri” menu item appears in native app context menus, letting users pipe highlighted content directly into a Siri conversation for processing.

Third-party AI extensions. iOS 27 introduces a Siri Extensions API that allows Claude, Gemini, ChatGPT, Grok, and Perplexity to function as alternative backends for specific surfaces including Writing Tools and Image Playground. Users can designate third-party models as default for those features, making Apple’s platform a distribution layer for all major AI labs simultaneously.

The Gemini Architecture

Apple evaluated OpenAI, Anthropic, and Perplexity before selecting Google. Under the deal, Google’s fully trained Gemini models handle complex queries routed through Apple’s Private Cloud Compute — data does not reach Google Cloud servers directly. Apple is simultaneously distilling Gemini into smaller on-device models optimised for Apple Silicon, and the next generation of Apple Foundation Models will use Gemini as the base architecture.

The arrangement is white-labelled: end users see Siri, not Gemini. For Google, $1B annually is secondary to the positioning — Gemini now powers the default AI backend on both Android and iOS, covering roughly 3.5 billion active smartphones globally.

What Failed

Federighi’s August 2025 all-hands was blunt: Apple had tried to bolt LLM capabilities onto the existing deterministic Siri command stack, and the hybrid failed to meet Apple’s quality bar. The personalised Siri features promised at WWDC 2024 were pushed entirely out of iOS 18. The decision to license Gemini rather than ship an inferior in-house model is the clearest signal yet that Apple’s internal AI capability remains a tier behind the frontier labs.

Numbers

• $1B: Apple’s annual Gemini licensing fee
• 2B+: Active Apple devices the new Siri will run on from autumn 2026
• 15 years: Time since Siri last had a dedicated standalone app
• June 8: WWDC 2026 keynote date

Timing and Risk

Apple has previously promised and deferred Siri upgrades: the features announced in June 2024 arrived more than a year late. If WWDC delivers the always-on agent as described, iOS 27 ships in autumn 2026. If Apple delays again, the narrative gap between Apple Intelligence and GPT-5.5, Claude Opus 4.7, and Gemini 3.1 widens further — with John Ternus, who took over as CEO in September from Tim Cook, inheriting the same credibility deficit Cook left behind on AI.

The extensions framework may matter more long-term than Siri itself. If Claude, ChatGPT, and Gemini can operate at iOS system level through first-class APIs, Apple becomes an AI distribution platform rather than an AI competitor. That outcome suits the labs, suits Apple’s services revenue model, and leaves consumers with more choice than the current default-model paradigm delivers.