Apple in talks with startup that shrinks AI models to run on an iPhone

Apple is currently engaged in discussions with a startup called PrismML regarding technology designed to shrink artificial intelligence models so they can run locally on iPhones. This development follows the release of Bonsai 27B by PrismML, which consists of 1-bit and ternary builds of Qwen3.6-27B. According to the reports, these specific builds are engineered to function on laptops and mobile phones. PrismML has explicitly confirmed that it is in talks with Apple about this model-shrinking technology, signaling a potential shift in how large-scale AI is deployed on consumer handheld devices. Coverage of these negotiations has been reported across several major business and technology outlets.

CNBC and qz.com both highlight that Apple is seeking ways to squeeze powerful AI models onto its iPhones. The technical specifics of the release are detailed by MarkTechPost and 9to5Mac, with the latter noting that PrismML claims Bonsai 27B is the first major AI model of its specific size that is fit for an iPhone. Additionally, AppleInsider reports on the confirmation from PrismML regarding the ongoing conversations between the startup and the tech giant. To understand the significance of these talks, one must look at the technical nature of the Bonsai 27B release. The model utilizes 1-bit and ternary builds based on Qwen3.6-27B, a method intended to reduce the computational and memory requirements of the AI.

This allows models that would typically require massive server infrastructure to operate on the limited hardware of a laptop or a smartphone. The goal is to enable powerful AI capabilities to run on-device, which would potentially reduce the need for cloud processing and improve the efficiency of AI tasks on the iPhone. Future developments will likely center on whether these talks result in a formal partnership or integration of PrismML technology into Apple's ecosystem. Observers will be watching for further updates on the compatibility of the Bonsai 27B model and whether other large-scale models will be subjected to similar shrinking processes. Coverage does not yet specify the exact terms of the discussions or a timeline for any potential software release, but the current focus remains on the technical viability of running the Qwen3.6-27B architecture on mobile hardware.