Local AI apps that finally make running an LLM on my phone worth it

A new wave of “local” AI apps is making on-device large language models practical on phones, according to a hands-on report by Bryan Wolfe. Wolfe previously tried a basic chat app in 2024 using a small 1B model, but it produced slow, unstable responses and was deleted after about a week. After driving through areas with weak or no signal around State College, Pennsylvania, he tested modern models and found performance far better. He highlights Qwen 3.5, Gemma 4, and Liquid AI’s LFM 2.5 running on phones, plus Apple’s on-device models made available to third-party apps via the Foundation Models framework. On an iPhone 17 Pro Max with iOS 27, five apps made his home screen, and one, LM Link, connects the phone to a Mac for heavier models using end-to-end encryption.