PrismML's Bonsai 27B squeezes 27.8 billion AI parameters into 3.9 GB, running locally on iPhone 17 Pro with 90-95% benchmark retention and no cloud needed.

PrismML says its compressed version of Alibaba’s Qwen model uses up to 15 times less memory, potentially advancing Apple’s AI push.

Apple is reportedly negotiating with PrismML to compress massive AI models for on-device iPhone use, with implications for privacy, computing, and crypto