Apple has held meetings with Prism ML, a startup that compressed Qwen 3.6 — Alibaba's 27-billion-parameter large language model — to run locally on an iPhone 17 Pro, according to The Information. Most mobile models operate with only a few billion active parameters, making this a significant leap in on-device intelligence. Prism ML plans to release its open-source version on July 14, claiming capabilities that include software development tasks.

The talks signal Apple's urgency to close the gap in generative AI without relying on cloud infrastructure. Running models of this scale locally would preserve privacy, reduce latency, and function offline — key differentiators as competitors push server-dependent features. The startup's approach could let Apple deploy advanced reasoning across its ecosystem without the data-center costs that burden rivals.

This pursuit aligns with Apple's largest known AI acquisition to date: Q.ai for a reported $2 billion. Simultaneously, the company partnered with Google to build the modern Siri AI slated for iOS 27. Prism ML's technology could accelerate that roadmap, letting Apple ship server-grade models on hardware it already sells.

If integration succeeds, consumers gain ChatGPT-level utility without sending prompts to external servers. For the industry, it pressures Android and PC makers to match on-device parameter counts or concede the privacy-performance narrative to Cupertino.