🤖Apple Rebuilds Siri on Google Gemini for WWDC 2026
TL;DR
At WWDC on June 8, Apple is set to unveil a rebuilt Siri trained off a ~1.2T-parameter Google Gemini model, with Apple reportedly paying Google about $1B a year. A smaller distilled version runs on-device, while heavy queries hit Google's Nvidia B200 servers.
At WWDC on June 8, Apple is set to unveil a rebuilt Siri trained off a ~1.2T-parameter Google Gemini model, with Apple reportedly paying Google about $1B a year. A smaller distilled version runs on-device, while heavy queries hit Google's Nvidia B200 servers.
Key Points
WWDC keynote lands Monday, June 8, 10am PT / 1pm ET
Revamped Siri leans on a custom model distilled from Google's ~1.2T-param Gemini
Apple reportedly pays Google around $1B per year for the model
Cloud queries run on Google's fleet of Nvidia Blackwell B200 chips
A standalone Siri app is expected to take on ChatGPT, Claude, and Gemini
Why It Matters
Apple effectively renting a frontier model from Google is an admission that its in-house AI fell behind, and it hands Gemini distribution to a billion-plus devices.
Quick Facts
Frequently Asked Questions
Why does this matter?
Apple effectively renting a frontier model from Google is an admission that its in-house AI fell behind, and it hands Gemini distribution to a billion-plus devices.
What happened?
At WWDC on June 8, Apple is set to unveil a rebuilt Siri trained off a ~1.2T-parameter Google Gemini model, with Apple reportedly paying Google about $1B a year. A smaller distilled version runs on-device, while heavy queries hit Google's Nvidia B200 servers.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 1,960 builders reading daily.