Apple Ships Long-Delayed Siri AI Overhaul at WWDC 2026, Built Partly on Gemini

The redesigned Siri retrieves online information, accesses personal context, understands speech/text/images, and acts across apps with onscreen awareness — capabilities Apple promised in 2024 but repeatedly delayed. Reporting confirms a roughly $1B/year arrangement with Google to power parts of the experience with Gemini, and the Swift LanguageModel protocol makes backends (Apple, Gemini, Claude) swappable via configuration — turning model choice into 'a setting' and reducing enterprise switching friction.

The rollout is gated: Siri AI runs only on newer Apple hardware, some features require an iCloud+ subscription, and regulatory restrictions will keep it out of Europe and China. New features are in developer testing now with a user beta later in the year. Bloomberg's hands-on described the assistant as belated but credible, framing it as the start of Apple's recovery from a major PR debacle rather than a leapfrog of rivals.

Competitively, leaning on Gemini is a pragmatic concession that Apple's own foundation models still trail frontier labs — a notable admission for a company that historically prizes vertical control. The configuration-as-model-selection design also hints at a future where Apple commoditizes the underlying LLM. Watch next: real-world latency and accuracy once the beta widens, and whether the EU/China gating pushes Apple toward on-device-only variants for restricted markets.