| Mobile Web

Cloud no longer needed? Ultra-large 400B-class LLM runs on iPhone in test

An experiment using Apple hardware and external storage has succeeded in running an ultra-large AI model on-device, drawing renewed attention to on-device AI. AI researcher Dan Woods (댄 우즈) ran a custom Qwen3.5-397B-A17B model on a MacBook Pro by storing weights in external flash memory and loading them only when needed. A similar test later ran on an iPhone 17 Pro at 0.7 tokens per second. Most code was written by Claude Opus 4.6.