Apple has held meetings with PrismML, a Caltech-linked startup working on extreme model compression, about ways its technology could help much larger AI models run directly on iPhones, according to a new report. The report follows PrismML’s claim that it compressed Alibaba’s open-source Qwen 3.6 model with 27 billion parameters to run entirely on an iPhone 17 Pro. That would be larger than Apple’s reported on-device AFM 3 Core Advanced model, which is said to have 20 billion parameters and power some iOS 27 features on newer devices. Apple has not announced a partnership with PrismML, and the meetings do not confirm that the startup’s technology will appear in future products. Better compression could give Apple more room to improve Siri, language tools, visual intelligence and offline features without sending every demanding request to the cloud. On-device AI has become one of Apple’s most urgent technical battles as the company tries to make iPhone intelligence faster, more private and less dependent on cloud servers.