Local LLMs on Arm laptops: actually usable, with caveats
We benchmarked four 8B-class models across three NPUs. The story isn’t about tokens per second.
Two years after the consumer AI boom, the device makers are stuck with the same product idea: a chatbot in a window. We spoke to engineers at six handset vendors to see how that’s about to change — and why it isn’t going to look like a chatbot at all.
We benchmarked four 8B-class models across three NPUs. The story isn’t about tokens per second.
Three big rings, two smart-glasses bets, and one company quietly killing it on chargers.
Battery life is finally where it should have been three generations ago.
If you bought the M3 model two years ago, sit this one out. If you held onto an M1 — read this first.
Better processor, calmer panel, identical remote. Worth the upgrade only from C2 or older.
Cursor, Windsurf, Copilot, Continue. We tracked how engineering teams actually pick.
“The interesting part of on-device AI isn’t the model. It’s the operating-system surface area that the model can finally touch without asking.” — from ‘Inside the quiet race to build a phone that thinks’