Apple MLX

MLX is Apple's open-source array framework purpose-built for Apple Silicon, designed for machine learning from basic numerical computations to running frontier models. Leveraging unified memory architecture, MLX eliminates CPU-GPU data copying while providing a familiar NumPy-like API with PyTorch and JAX influences. Released in December 2023 and featured at WWDC 2025, MLX supports running massive models like DeepSeek AI (670B parameters) on M3 Ultra, achieving faster-than-reading-speed text generation with efficient 4-bit quantization. Available in Python, Swift, C++, and C, MLX is MIT-licensed and optimized for Apple's unique hardware capabilities.