Ollama flips Apple Silicon to MLX by default in 0.40.0 rc0
The useful change today is in Ollama’s v0.40.0-rc0 release: supported models now run on MLX on Apple Silicon by default. If you are buying a Mac because you want the shortest path from unboxing to local inference, that matters more than a small price wiggle. It removes one more runtime choice from the setup path on Apple’s machines.
That does not change memory math, and it does not mean every model suddenly fits on every Mac. It does make the default local stack friendlier on the boxes people actually cross-shop for home AI right now, especially the Mac mini M5 Pro 64GB, the Mac Studio M5 Max 128GB, and the older Mac Studio M4 Max 128GB. If your real alternative is an RTX 5090 system, the buying question is still memory capacity versus CUDA software breadth, but the Apple side is a little easier to justify when convenience is part of the brief. The fast next click is the live Mac mini M5 Pro vs Mac Studio M5 Max comparison.
There are two caveats. First, this is still an RC, and the release notes say Ollama is still testing and enabling additional models during the release candidate period. Second, Apple’s current Mac Studio buy page still says the 512GB memory option for M5 Ultra is coming late October, so the highest-memory Mac in our catalog is not yet a normal click-to-buy option. For a buyer today, the practical takeaway is simple: if you want the easiest current Mac path for Ollama, the tracked M5 Pro and M5 Max pages matter right now, and the M5 Ultra remains the one to watch.