Ollama adds MLX boost on Macs

This is a meaningful Mac-native upgrade, not just a benchmark victory. Ollama is leaning into Apple’s hardware model instead of fighting it, which makes local inference feel more practical for real work.

• –Apple Silicon’s unified memory is the real unlock here: less VRAM-style friction, better fit for larger local models, and a smoother path for multitasking on a laptop or mini
• –Ollama says the preview is substantially faster on Apple Silicon, with the biggest gains aimed at agentic and coding workloads where latency and responsiveness matter
• –The update doesn’t erase the high end: heavy serving, training, and throughput-sensitive deployments still belong on NVIDIA hardware
• –Community momentum around Apple Silicon benchmarks and quantization improvements suggests the software stack is improving fast enough to change buying and workflow decisions
• –For many developers, the practical win is not peak speed but turning a Mac from “good enough to test” into “good enough to keep using”