Apple M5 MLX: 70B LLMs Go Portable in 2026

Benchmarks of Apple's M5 chip with MLX show 153GB/s bandwidth, 40 Neural Accelerators, and 20-50% faster inference than llama.cpp for 70B models on a laptop.