Apple dropped the 2018 Mac mini from macOS Tahoe, then dropped every Intel Mac from macOS 27. On 14 October 2026 we checked what that machine’s software update feed was actually offering. It offered Safari 27.0 — the same browser generation that shipped alongside the release the machine cannot install.
Being cut off is not one thing. On the same day, a Mac on Sonoma was offered nothing, our Mac mini on Sequoia was offered a current-generation browser, and a Mac on Tahoe was offered both a browser and a system update. Where your machine sits in that order decides whether it still gets the modern web engine, and almost nothing written about the Intel cutoff makes that distinction.
We own the machine in the middle. Here is what it gets, and what it can still do.
The machine
A Mac mini (2018) with a 6-core Intel Core i5-8500B at 3.0 GHz and 32 GB of RAM, running macOS Sequoia 15.7.7 (build 24G720). It is not a museum piece sitting in a drawer. It runs browser automation and background jobs continuously, which matters for the numbers below.
This machine was never on the Tahoe compatibility list. That list contained four Intel models — the 2019 Mac Pro, the 2019 16-inch MacBook Pro, the 4-port 2020 13-inch MacBook Pro, and the 2020 27-inch iMac. A 2018 Mac mini is not among them, so Sequoia 15 is the end of its upgrade path. macOS 27 then removed Intel support entirely, which changes nothing for this machine because it had already stopped one release earlier.
What the update feed actually offered
We did not read an announcement. We asked the machine, with softwareupdate -l, and logged the answer:
| Item | Offered on 14 Oct 2026 |
|---|---|
Safari27.0SequoiaAuto-27.0 |
Yes |
macOS Sequoia 15.8.1-24H32 |
Yes |
Two things are worth separating here.
The system update is a minor-version bump, not just a patch. The machine is on 15.7.7 and is being offered 15.8.1. Apple moved this legacy line from 15.7.x to 15.8.x rather than continuing to increment the patch digit.
The browser is current-generation. Safari 27.0 was released on 14 September 2026, the same day as macOS 27. Apple’s own release notes date it to that day. Our machine — excluded from the last two macOS releases — is offered that exact build.
That is the part that surprised us. The usual mental model of a cutoff Mac is a machine frozen in place, slowly falling behind on web standards until sites break. For this machine, in October 2026, that model is wrong. The operating system is frozen. The browser is not.
It is also not universal. Apple’s own security releases page lists Safari 27 as available for “macOS Sequoia and macOS Tahoe” — Sonoma is absent. On that same date, 14 September 2026, the page carries no macOS Sonoma entry of any kind. Machines one step further back than ours got neither a system update nor a browser update that day.
The cutoff has levels, then, and they do not line up with the headline. The headline cutoff is about which Macs can install the new OS. The cutoff that decides whether your browser stays current runs somewhere else entirely — between Sonoma and Sequoia — and our machine is on the right side of it.
How much machine is left
A browser that can render the modern web is only useful if the hardware can run it. So we measured.
Both machines were tested with the same command, the same binary, and the same block size: /usr/bin/openssl speed -evp sha256, reading the 8192-byte column, in units of 1000 bytes per second. We used /usr/bin/openssl deliberately — it is LibreSSL 3.3.6 and ships on every Mac, so any reader can reproduce this without installing anything.
| Mac mini (2018, i5-8500B) | MacBook Pro (M5 Pro) | |
|---|---|---|
| Single thread | 368,539 | 3,384,128 |
| All cores | 1,216,036 (6 threads) | 49,820,010 (18 threads) |
Higher is better. The single-thread gap is 9.18x.
There is a cleaner way to say the same thing. One M5 Pro core produces 3,384,128. Six Mac mini cores, if they scaled perfectly — which no processor does — would produce 368,539 x 6 = 2,211,234. A single modern core beats all six older cores by about 1.5x, even after handing the older machine an impossible best case.
For a sense of where this sits against the machines we have measured before, our M1 Pro to M5 Pro comparison covers the Apple silicon end of the same spread.
What these numbers are not
They are lower bounds, and we want to be exact about why.
The Mac mini was running its normal workload during every measurement — headless Chromium, a Node process, a Supabase MCP server, depending on the run. We did not stop that work, because that is the machine’s real condition and stopping it would measure a situation that never occurs.
The consequence shows up in the spread. Across four sessions, the all-core result ranged from 770,480 to 1,216,036 — a 37% swing on the same hardware in the same afternoon. We report the highest observation, because contention can only take performance away: no session can exceed what the chip can do, so the maximum is the best available estimate of the floor, and it rises toward the truth as sessions accumulate. This is the same problem we wrote about in why benchmark results disagree with each other, arriving from the opposite direction.
Single-thread behaved far better, landing between 351,959 and 379,752 across the same four sessions — a 7.4% spread. That is expected: a single-threaded test needs one free core, and this machine always had three or more idle. The 9.18x figure rests on the stable measurement, not the volatile one.
We did not measure this machine’s ceiling. A quiet 2018 Mac mini would score higher than 1,216,036 on six threads. We do not know how much higher, and we are not going to estimate it.
One more limit worth stating. We confirmed on the M5 Pro that LibreSSL and Homebrew’s OpenSSL 3.6.3 agree closely — 0.4% apart on single thread, 3.5% on all cores — which is why we are comfortable telling readers to use the built-in binary. We did not run that same cross-check on Intel, so we cannot promise the two libraries agree there.
The throttling question, answered directly
Old machines are assumed to be thermally compromised. This one is not, at least not now.
pmset -g therm reports CPU_Speed_Limit, a counter that exists only on Intel Macs — Apple silicon does not populate it, which is why our M-series sustained-load testing had to infer thermal behaviour from repeated timings instead of reading it. On the Mac mini we can simply read the number.
It reported 100% in every session, before and after load, including a session where the machine was already 59.6% busy before our test began. No thermal limiting was applied. Whatever is slowing this machine down relative to a modern Mac, it is the 2018 silicon, not heat.
What we would tell someone with one of these
If you have an Intel Mac on Sequoia, the browser is not your problem yet. You are being offered the same Safari generation as machines running the current OS, and that is the component that decides whether the web works.
What you are losing is the operating system’s feature line, and roughly 9x of single-core performance against current hardware. Whether that matters depends entirely on what you ask the machine to do. Ours spends its days running headless browser jobs and background automation, where the work is patient and the clock is not. For that, it is still a working computer.
We will keep logging this machine’s update feed on a schedule. The interesting measurement is not today’s snapshot — it is what the feed offers six months from now, and whether 15.8.1 makes this machine measurably slower or faster once it installs. We have four sessions of before-data already.
Measurements taken 15 October 2026 on hardware we own. Raw figures and session-by-session spread are listed above. Not investment advice.