Meetings with PrismML have been held with Apple, discussing ways to adopt the technology from the startup to be able to run significantly larger AI models. Qwen 3.6, the Alibaba open-source LLM, has been shrunk down by PrismML to be able to run on the iPhone 17 Pro, featuring 27 billion parameters, bigger than the AFM 3 Core Advanced on-device model that has 20 billion parameters.
A larger model being run on the iPhone directly could enable more features of Apple Intelligence to be integrated and run on the device directly instead of running on the Apple Private Cloud Computer servers, which could help lessen the costs for Apple and provide better privacy for the user.
Qwen 3.6 and its parameters, compared to the AFM 3 Core Advanced, are able to be active at the same time.
