New AI Model Operates Natively on 12GB Memory iPhones
The Bonsai 27B AI model, fine-tuned from Qwen 3.6 27B, can now run natively on iPhones with 12GB of RAM while retaining 90% of its intelligence.
Developer PrismML announced the release of the Bonsai 27B AI model, engineered for on-device operation. This model is derived from the Qwen 3.6 27B architecture but significantly optimized for reduced memory consumption.
Two versions of Bonsai 27B are available: a 3-bit Ternary version requiring 5.9 GB, targeted for laptop-level performance, and a 1-bit version utilizing 3.9 GB, specifically designed for smartphone footprints.
This development allows advanced AI capabilities, including multi-step reasoning and long-context workflows, to operate directly on a device without reliance on cloud servers or high-performance hardware.
The release marks a significant advancement in accessible, local AI processing, potentially enabling wider integration into consumer electronics.