PrismML Shrinks 27B AI Model for iPhone 17 Pro

PrismML has successfully compressed a 27 billion-parameter AI model to run directly on an iPhone 17 Pro. This Caltech spinout shrank Alibaba's Qwen 3.6 model from 54 gigabytes to under 4 gigabytes. Here's the thing: every single parameter remains active. The company's new Bonsai model family offers up to a 14 times smaller memory footprint and 8 times faster speeds. What's interesting is it achieves this without removing any core capabilities. Apple is reportedly discussing integrating this technology into its future AI products. This could allow Apple to surpass its current on-device AI limit of 20 billion parameters without redesigning its mobile chip. This means more powerful AI could run directly on your phone, keeping data private and potentially saving energy.