At WWDC 2026, Apple announced its third-generation Apple Foundation Models (AFM 3), including a 20-billion-parameter model that runs offline on an iPhone. Announced June 8, 2026, AFM 3 has five models: on-device and server. The on-device AFM 3 Core Advanced has 20 billion parameters using sparse architecture, activating only 1 to 4 billion per request. On-device models also include AFM 3 Core with ~3 billion parameters. Server-based models via Private Cloud Compute include AFM 3 Cloud, ADM 3 Cloud (Image), and AFM 3 Cloud Pro. For text performance, AFM 3 Cloud jumped from 8.7% user preference (2025 model) to 64.7%. On-device, AFM 3 Core moved from 23.3% to 45.6% preference. Overall satisfaction improved by 36%, and instruction-following is up 21%. Dictation quality preference rose from 17.6% to 44.7%. Expressive voice now scores a Mean Opinion Score of 4.15 to 4.24, up 0.28 to 0.42 points. AFM 3 Cloud Pro showed a 14% gain over the standard cloud model for image tasks. AFM 3 does not contain “a drop of Gemini” for live processing; Gemini was used only during