Apple unveiled a rebuilt Siri at its WWDC keynote on Monday powered by a custom, roughly 1.2-trillion-parameter Google Gemini model that the company is licensing for about $1 billion a year, according to Bloomberg’s Mark Gurman. The arrangement is the clearest admission yet that Apple’s frontier-model ambitions have been quietly outsourced.

The new Siri architecture is structured in three layers, per Business Standard citing Bloomberg: a query planner, a knowledge retrieval engine, and a summarisation layer, with Gemini handling planning and summarisation while Apple’s on-device models keep local tasks local. Tom’s Guide, citing Bloomberg, says Siri also gets a standalone app with a Dynamic Island presence, a chatbot-style “Ask Siri” bar, paperclip attachments for images and PDFs, and rich result cards. The personal-context features promised two years ago, reaching across emails, photos, calendar entries, and files, finally arrive.

Apple is also opening the floor. A new Extensions framework, accessible through the Apple Intelligence settings panel, lets users route queries to Anthropic’s Claude, Google’s Gemini, or OpenAI’s ChatGPT as third-party assistants. It’s a quiet capitulation to the post-default-search world, where the assistant slot is the next contested layer.

iOS 27, iPadOS 27, macOS 27, tvOS 27, watchOS 27, and visionOS 27 all ship together, per MacRumors, with developer betas seeding after the keynote, public betas in July, and a full public release in the fall. macOS 27 drops Intel Macs.

The keynote is widely expected to be Tim Cook’s last as chief executive, with hardware chief John Ternus the presumed successor. Cook presided over Siri’s debut in 2011. He now presides over its replacement by someone else’s model.

Sources