AI Models

Small models, big shift: on-device AI assistants go mainstream

Sub-10-billion-parameter models now handle most everyday assistant tasks entirely on phones and laptops, cutting latency, cost and privacy exposure at once.

By Maya Okafor, Senior AI Correspondent — SAN FRANCISCO

SAN FRANCISCO — The most consequential AI models of the year may be the smallest. Compact language models under ten billion parameters now resolve the bulk of everyday assistant requests — summarising, drafting, translating, controlling apps — entirely on the device in your hand.

Device makers say distillation from frontier teachers, aggressive quantisation and purpose-built neural silicon have lifted on-device quality past the threshold where most users notice no difference — while latency drops to milliseconds and sensitive data never leaves the phone.

Analysts note the shift quietly redraws the competitive map: control of the operating system, not the largest datacenter, decides which assistant a billion people meet first.

Enable JavaScript to read the full story on Neural Daily News.