Aug. 25, 2026

What a $100 Intel Mac mini is still good for

Apple Silicon minis hold most of their value used. The Intel minis they replaced sell for $72 to $625 — and for a surprising number of home-server jobs, the old machine does the work. A use-case guide, with the honest limits.

A $100 Intel Mac mini is still useful as a file server, Time Machine target, direct-play media server or iMessage bridge. A used M2 mini asks about $610 — more than it cost new in 2023 — while the Intel models it replaced start at $72 in our Aug. 24-25 market snapshot.

The tradeoff is security and speed. Most Intel minis need Linux or a patched macOS for current security updates, and none is a sensible local AI machine.

What the old minis cost

One correction to a common assumption before reading further: there was no quad-core 2014 mini. The 2014 line is dual-core only, and its best i7 benchmarks below the 2012 quad — PassMark 3,054 against 5,158. The $202 machine from 2012 is faster than the $165 machine from 2014.

Which Intel Mac mini fits each home-server job?

The iMessage bridge deserves its line of fame: it is the job that specifically requires a Mac — any Mac — and nearly any Mac will do. An agent setup like OpenClaw’s iMessage channel needs a machine signed into Messages plus Node 22, which means a 2018 mini on its stock OS, or a 2012/2014 running a patched newer macOS. A plain BlueBubbles bridge runs happily on the 2012’s native Catalina. Either way the hardware is idle-load; the $72 machine clears it.

The 2018 i3 is the sleeper in the table. At $200 it carries the same hardware HEVC engine as its $625 six-core sibling, which is the part that matters for a 4K Plex box.

The same yardstick, both eras

Score the Intel minis the way we scored the Apple Silicon used market: benchmark points per dollar of median ask. One wrinkle: the old machines don’t appear on the Geekbench 7 charts our Silicon table used, so this table runs on Geekbench 6 per-model averages, all from EveryMac’s independently calculated figures — one source, one scale, both eras. The numbers aren’t comparable to the GB7 table; the ratios are the point.

The finding mirrors the Silicon table’s, and extends it: the used market prices raw compute at a strikingly consistent rate across 15 years of hardware. The 2014 i5 and 2018 i3 sit at 17-18 points per dollar; the M1 and M2 sit at 16-17. Nobody planned that. Thousands of individual sellers converged on roughly one price for a unit of multi-core throughput, whatever the badge says.

Two honest reads follow. The Intel minis are not a compute bargain — the efficient market took care of that — and the worst deal on the whole board is the 2018 six-core i7 at 8.9, its price propped up by being the last Intel mini that still gets security patches. What the Intel machines offer is a different thing: a low absolute price for jobs where compute is irrelevant. A $72 iMessage bridge doesn’t care about points per dollar. And the single-core column is the fine print on every Intel row: the 2014 i5’s 887 is a 2011-class experience at the keyboard, less than a quarter of the M4’s 3,791.

Are old Intel Mac minis still secure?

On its last official macOS release, every model below the 2018 runs an unpatched operating system. For a LAN Plex box or a Time Machine target behind your router, that is a risk many people reasonably accept. For anything exposed to the internet, it isn’t. OpenCore Legacy Patcher can put current macOS Sequoia on the 2012 and — most cleanly — the 2014, which restores patch coverage at the cost of an enthusiast-grade install; the 2011’s version of that path runs through graphics patches rough enough that we wouldn’t call it smooth. Linux is the other honest route to a patched old mini, and for most of the server jobs above it is the better one.

Can an Intel Mac mini run local AI?

No. None of these machines can do local AI work in any useful way. The 2011 and 2012 predate AVX2, the instruction set most modern inference runtimes assume, so common builds simply won’t run; the 2014 has two cores; the 2018 has no neural hardware and no unified memory. We found no citable benchmark of a current 7B-class model on this hardware — our estimate from the instruction set and memory bandwidth is low single-digit tokens per second at best, and we may put a number on it ourselves. If local models are the job, that is what the Apple Silicon used market is for — and it’s priced accordingly.

About the data: Prices are median asking prices for live listings observed by our crawler Aug. 24-25, with scam-shaped listings excluded; asking prices are not completed sales. macOS support and requirement facts come from Apple's support pages, project documentation and PassMark's published CPU scores, linked in place. The Sequoia end-of-support estimate is a projection from Apple's usual three-year pattern, not an Apple commitment.