Apple overhauls Siri at WWDC 2026 with screen awareness and new on-device models

Apple used WWDC 2026 to relaunch Apple Intelligence around a rebuilt, screen-aware Siri capable of acting on what's currently displayed, marking its most significant assistant overhaul in years. Reports indicate the new Siri leans on foundation models developed with Google Gemini, running on NVIDIA hardware via Google Cloud, alongside Apple's own on-device stack and Private Cloud Compute for privacy-sensitive tasks. Siri entered developer beta on June 9, with a wider US English beta expected later in 2026.

Technically, Apple introduced two new foundation models: AFM 3 Core, a 3-billion-parameter dense model, and AFM 3 Core Advanced, a 20-billion-parameter sparse, multimodal model. Visual Intelligence expands across the ecosystem — into Wallet (scanning receipts and splitting bills via Apple Cash), Maps (Local Lists and improved Flyover), Music, Photos, Safari, and Shortcuts — with image generation and automated task execution woven throughout.

The strategic subtext is redemption: Apple is seeking to recover credibility after a $250M settlement tied to earlier Apple Intelligence shortfalls and a Siri that lagged rivals. Leaning on Gemini is a pragmatic admission that Apple's in-house models trail the frontier, even as it markets on-device privacy as a differentiator. The competitive question is whether a screen-aware, action-taking Siri can close the gap with Google's Gemini agents and OpenAI's ChatGPT platform — or whether dependence on Google's models simply cedes the hardest AI work to a rival. Watch beta feedback and whether the wider rollout slips again.