Local AI at Every Memory Size: the Cheat Sheet

How much memory you need for local AI depends on what you want it to do, and the answer changes a lot between sizes. A 16 GB Mac has about 3 GB left for the AI while your browser and mail are open. A 128 GB Mac Studio runs a model a bit under half as smart as today’s best paid AI, with room to spare.

This sheet puts every size from 16 GB to 1 TB side by side, with US prices from late October 2026:

• The ladder at a glance: the cheapest way in at each size, the Mac way in, the model to run, how much it holds and how smart it is.
• Every way in, cheapest to most expensive: Macs, graphics cards, AMD boxes, DGX Sparks and rented cloud chips.
• How much of the memory the AI really gets on each machine, with your apps open and closed.
• What fits: each model’s context window on each machine, and whether it’s enough for agents.
• How fast it writes, from the published runs of owners and testers. Our estimates are marked as estimates.
• What I’d buy at each budget, and six things to know before you spend.

Scores are the Artificial Analysis Intelligence Index v5 (29 October 2026), as a share of Claude Opus 5.5, one of today’s two best models. Models it hasn’t tested yet are estimated and marked.