Apple is reportedly interested in acquiring or partnering with PrismML, a startup that has developed technology to run massive language models directly on smartphones. According to 9to5Mac, PrismML claims to have successfully executed Qwen 3.6—a model with 27 billion parameters—on an iPhone 17 Pro.
This achievement represents a significant leap in on-device AI capability. Typically, smartphones run language models with around 3 billion parameters on average. PrismML's approach delivers roughly 800% more capacity, bringing genuinely powerful AI inference to mobile devices without relying on cloud servers. For users, this means faster responses, better privacy, and AI features that work offline.
If Apple moves forward with the technology, it could reshape how iPhone users interact with AI. Running large models locally eliminates latency and dependency on network connectivity, while keeping sensitive data on the device rather than sending it to remote servers.
