Apple has held meetings with the startup PrismML to explore ways to run significantly larger artificial intelligence models locally on iPhones, according to a report from The Information. The initiative aims to boost on-device processing for Apple Intelligence features while strengthening user privacy and cutting cloud-related expenses.
PrismML shrinks Alibaba's Qwen 3.6 to run entirely on iPhone 17 Pro
The startup successfully compressed Alibaba's open-source large language model Qwen 3.6 so that it operates fully on an iPhone 17 Pro. The model contains 27 billion parameters, surpassing Apple's own AFM 3 Core Advanced model which has 20 billion parameters. Apple's model powers iOS 27 enhancements such as Siri AI's more expressive voices and improved systemwide dictation on iPhone 17 Pro and iPhone Air models. However, Apple uses a sparse architecture where only 1 to 4 billion parameters are active at any time. In contrast, PrismML's solution keeps all 27 billion parameters active simultaneously, offering greater computational capacity.
Sponsored Protocol
Privacy and cost benefits through local processing
Running larger AI models directly on the device would allow many Apple Intelligence features to function without sending data to Apple's Private Cloud Compute servers. This reduces reliance on external infrastructure, lowers operational costs for Apple, and significantly enhances user privacy. In a landscape where incidents like Madison Square Garden spying on celebrities with a secret risk database highlight privacy vulnerabilities, keeping sensitive data on-device is a crucial improvement.
Comparison with other AI voice solutions and future outlook
Apple's move follows OpenAI's rollout of GPT-Live duplex voice, which brought real-time translation and visual cards to ChatGPT. While ChatGPT relies on cloud processing, Apple is aiming to keep computation local, a choice that could offer faster response times and stronger privacy. If the partnership with PrismML succeeds, future iPhones could host even more complex models, expanding Siri's capabilities and other AI functions without compromising security. The underlying technology of large language models, as described on Wikipedia, is advancing rapidly, and Apple appears determined to leverage these innovations to keep its ecosystem at the forefront.
Sponsored Protocol
Source: https://www.macrumors.com/2026/07/09/apple-prismml-larger-on-device-ai-models
