PrismML claims it successfully operated a 27-billion-parameter Qwen 3.6 model on an iPhone 17 Pro. This achievement marks the largest AI model ever to run on a smartphone.
The Khosla Ventures-backed startup compressed the model from 54 GB to under 4 GB. The technology maintains all parameters in an active state during local execution.
Running models locally enhances user privacy and improves response times. The process also significantly reduces cloud computing costs.
Apple has reportedly held talks with PrismML regarding the technology. The discussions signal Apple's intent to advance on-device AI capabilities for future product lines.