Blog.

Notes from the team on compression, sparse silicon, and shipping AI on real devices.

LATEST
JULY 16, 2026

Say the Word, at 60 KB

We pruned a voice command model from 500 KB to 60 KB while keeping a 95% detection rate at 0.25 false alerts per day.

Scott Reid, Debayan Ghosh, and Jack Ferrari

Read the post
ALL POSTS · 5

GET UPDATES

New posts and product news, in your inbox.

A short note when there is something worth reading. No more than a few times a quarter.

Announcing_

Talk to the team.

Briefings, interviews, and technical deep dives for press and analysts.

Contact the press team