en.Wedoany.com Reported - PrismML has successfully ported Alibaba's open-source 27-billion-parameter large model Qwen 3.6 to the iPhone 17 Pro. According to IT news service The Information, the model is capable of handling complex chat responses, reasoning tasks, and fully autonomous coding agents. Starting next Tuesday, the model will be made available to the public.

PrismML is a spin-off from the California Institute of Technology (Caltech), and its head, Baba Hassibi, also serves as a professor of electrical engineering at the university. The company claims to have compressed Qwen 3.6 using "mathematical tricks," surpassing conventional quantization methods. This technique is said to eliminate typical quality loss, reducing Qwen 3.6 from 54 GB to 4 GB. The relevant patents are held by Caltech.

PrismML predicts that within about three years, up to 95% of required AI computing power will be processed locally on smartphones, laptops, and even home appliances. Baba Hassibi believes that only 5% of high-end demand will go to the cloud at that time, stating, "I believe this is the direction people see moving forward." For training, data centers will remain necessary for the foreseeable future. Notably, Apple previously introduced a technology for AFM Core Advanced that addresses RAM issues for 20-billion-parameter large models by using flash memory to load only the truly needed parts. Apple is currently seeking AI companies in this field for potential acquisition. According to The Information, Apple has held preliminary discussions regarding the use of this technology.