Laya (OS Jev) demonstrates real-time AI inference on Apple silicon, achieving 45 decisions per second using Mac M4 CoreML. This performance milestone showcases the potential of offline large language models on consumer hardware, eliminating the need for cloud dependencies. The benchmark highlights significant optimization for Apple's neural engine, enabling responsive AI applications directly on desktop devices.

This development marks a shift toward privacy-focused, locally processed AI workflows. The achievement comes as major tech companies race to optimize models for on-device execution. Laya's success on M4 silicon suggests a viable path for developers seeking high-performance AI without internet connectivity requirements.