Devices · 2h ago
Apple and Google Ship On-Device AI That Finally Works Offline
Apple's new 20B-parameter on-device model activates only 1-4B parameters per request, running entirely offline. Google's Gemma 4 uses similar sparse activation to keep active memory low. This shift eliminates per-token cloud costs, enabling private, free AI agents on existing hardware.
Meridian48 take
The real breakthrough isn't raw parameter count but the economic and architectural shift to sparse activation, making on-device AI genuinely practical for the first time.
on-device-aiapple-google-models