If you are on an M4 Pro, skip this generation. Across every test we ran, the M5 Pro was between 6% and 9% faster — a gap you will not feel. If you are on an M1 Pro, the answer depends entirely on what you do with the machine: we measured anywhere from 1.40x to 2.38x, and the difference between those two numbers is the whole decision. Updated October 2026: on work that uses every core the top of that range is 3.47x, and the advice to M4 Pro owners is narrower than the line above — see the update directly below.
We tested three MacBook Pros we own, on the same day, with the same workloads.
Update, October 2026: one of our original rows was wrong, and it changes the advice
Two things have happened since we published this. All three machines are now on the same macOS, which lets us settle a limitation we flagged at the bottom of this article. And we found a fault in our own method that matters more than that.
The “all-core” row was not an all-core test
The all-core compression row below uses zstd -T0, which asks the tool to use every thread. We checked what it actually does, and on the 18-core M5 Pro it only keeps about 31% of the machine busy — the job spends the rest of its time waiting on the disk rather than computing. The number is a correct measurement of that compression job. It is not a measurement of what the chip can do with every core working.
So we re-ran all three on a workload that genuinely saturates every logical core, confirmed at over 97% occupancy, eight consecutive rounds each, same afternoon, all on macOS 27.0.1.
| Chip | Year | Cores | Every core working | Spread | Per core |
|---|---|---|---|---|---|
| M1 Pro | 2021 | 8 | 16.3M | 3.2% | 2.04M |
| M4 Pro | 2024 | 14 | 36.6M | 0.6% | 2.61M |
| M5 Pro | 2026 | 18 | 56.6M | 1.6% | 3.14M |
None of the three throttled. Every round on every machine reported a CPU speed limit of 100%, and the 2021 machine finished its eighth round 1.0% faster than its first.
What this does to the M4 Pro advice
The opening line of this article says that if you are on an M4 Pro you should skip this generation, because every test we ran put the M5 Pro 6% to 9% ahead. On work that uses every core, it is 55% ahead.
That is not a small correction, and it has a clear cause: a test that only loads a third of the machine cannot see the difference between 14 cores and 18. The old row was measuring the disk as much as the processor.
The revised advice is narrower than the original but more useful. If the jobs you wait on pin every core — compiling, rendering, bulk encoding, many containers at once — the M4 Pro to M5 Pro jump is real and you should not dismiss it. If they do not, the original 6% to 9% still describes your life, and skipping the generation is still right.
Where the four-year gap actually comes from
With every core working, the M5 Pro does 3.47 times the work of the M1 Pro. That is not one improvement. It is two, multiplied:
| Total | From more cores | From faster cores | |
|---|---|---|---|
| M1 Pro → M4 Pro | 2.25x | 1.75x | 1.28x |
| M4 Pro → M5 Pro | 1.55x | 1.29x | 1.20x |
| M1 Pro → M5 Pro | 3.47x | 2.25x | 1.54x |
In both generations Apple added cores faster than it made them quicker. That is the most useful thing in this table, because you only collect the 3.47x if your work uses every core at once. Everything else collects 1.54x — and a browser, a text editor and most of what a laptop does in an ordinary hour runs on a few cores.
It also explains why the range in the original table is so wide. Jobs that spread across cores sit at the top of it; jobs that do not sit at the bottom. The spread is not measurement noise. It is the shape of the upgrade.
One condition worth naming: each machine ran on its own supplied charger, so the wattages differed. None of them reported a power or thermal limit at any point, but the chargers were not identical. The September figures below are unchanged.
The numbers
| M5 Pro 16″ | M4 Pro 14″ | M1 Pro 14″ | |
|---|---|---|---|
| Single-core compression | 1.75s | 1.90s | 3.05s |
| All-core compression | 0.39s | 0.42s | 0.85s |
| CPU video encode (x264) | 3.60s | not tested | 8.58s |
| Hardware video encode | 3.34s | not tested | 4.68s |
| Browser responsiveness | 55.0 | 51.2 | 35.6 |
| Work done in 8 minutes | 1,215 | 1,148 | 562 |
Times are seconds, lower is better. Browser responsiveness is a Speedometer 3.1 score, higher is better. The last row counts how many identical compression passes each machine finished during a continuous eight-minute load.
One generation is not a reason to buy
M5 Pro against M4 Pro, every measurement we have:
| Test | Gap |
|---|---|
| Single-core | 1.09x |
| All-core | 1.08x |
| Sustained throughput | 1.06x |
| Browser | 1.07x |
Six to nine percent. Our run-to-run variation on these machines was 1.5% to 7.8%, so the gap is real and sits above our noise — but a 9% difference on a task that takes ten seconds saves you under a second. Nobody notices that. If you own an M4 Pro, there is no performance argument for replacing it, and we are not going to invent one.
Four generations is a different question, and it has no single answer
M5 Pro against M1 Pro, the same measurements:
| Test | Gap |
|---|---|
| Hardware video encode | 1.40x |
| Browser responsiveness | 1.54x |
| Single-core | 1.74x |
| Sustained throughput | 2.16x |
| All-core | 2.18x |
| CPU video encode | 2.38x |
The spread between the smallest and largest gap is 70 percentage points. This is the part most comparisons flatten into one number, and flattening it gives the wrong answer to half the people reading.
If your heavy work runs on the hardware video encoder — exporting footage in an editor that uses it — four years of silicon bought you 1.40x. That is the weakest result in our entire test set, and it is worth knowing before you spend money.
If your heavy work is CPU-bound — software encoding, compilation, anything that saturates cores — the same four years bought you 2.38x, and the machine finished 2.16x as much work during a sustained eight-minute load. That is a different machine in daily use.
Browser work sits in between at 1.54x, which is where most people actually live.
The core counts do not line up the way you would expect
Open Activity Monitor on each machine and the core labels tell their own story:
| Cores | Breakdown | |
|---|---|---|
| M1 Pro 14″ | 8 | 6 Performance + 2 Efficiency |
| M4 Pro 14″ | 14 | 10 Performance + 4 Efficiency |
| M5 Pro 16″ | 18 | 12 Performance + 6 Super |
The M5 Pro has no cores labelled Efficiency. What macOS used to call efficiency cores are now labelled Performance, and a tier above them is labelled Super — a name Apple’s own specification pages do not use. You can read it yourself with sysctl hw.perflevel0.name.
We are reporting the labels and the counts, which are verifiable. We have not yet measured what a Super core does differently, and until we pin work to specific cores we are not going to guess.
Who should move
Upgrade from M1 Pro if: your work is CPU-bound, or you regularly run loads long enough that throughput matters more than burst speed. The sustained numbers are where the M1 Pro falls furthest behind — 562 passes against 1,215, and on a load that uses every core the M5 Pro does 3.47 times the work. We worked that decision through on its own in should you replace an M1 Pro.
Stay on M1 Pro if: your heavy lifting goes through the hardware video encoder, or your work is mostly browser and text. A 1.40x gain on the work you actually do is not worth a four-figure purchase, and the M1 Pro held its performance under load as well as the newer machines did. For how it compares against everything else in the house on browser work, see five devices ranked by browser speed.
Stay on M4 Pro: unless your jobs pin every core. On everything else we measured the M5 Pro is 6% to 9% ahead, which is not a reason to spend. On all-core work it is 55% ahead, which might be. If you are weighing a new M5 anyway, we also looked at the problems being reported with it.
Check your own machine
Every figure here came from tools already on your Mac. The core tiers:
sysctl -a | grep perflevel | grep name
And an all-core run you can compare against the October table above. It needs no file and nothing installed, and unlike the compression command it keeps every core busy:
openssl speed -evp sha256 -multi $(sysctl -n hw.ncpu)
Take the figure from the last column of the final row: 16.3M, 36.6M and 56.6M are what our three machines produce. For the single-core compression row in the September table instead, use:
zstd -12 -f -q -T1 -o /tmp/out.zst /tmp/bigfile.bin
Run each three times and take the median. Close your browser first — leaving ours open cost us 20% and produced a result that was simply wrong.
Common questions
How much faster is an M5 Pro than an M1 Pro?
Between 1.40x and 3.47x, and which end you get depends entirely on the work. On a load that uses every core it is 3.47x. On work that runs on one or two cores — a browser, a text editor — it is 1.54x. Through the hardware video encoder it is 1.40x. Picking the row that matches your work matters more than the headline.
Is it worth upgrading from an M4 Pro to an M5 Pro?
Only if your jobs pin every core. On everything else we measured the M5 Pro is 6% to 9% ahead, which you will not feel. On all-core work it is 55% ahead, because 18 cores against 14 is a real difference that lighter tests cannot see.
Do the extra cores actually translate into speed?
More than the cores themselves do. From M1 Pro to M5 Pro, 2.25x of the gain comes from having more cores and only 1.54x from each core being quicker. In both generations Apple added cores faster than it made them faster — which is why the benefit disappears on work that cannot use them.
Does the M1 Pro throttle under sustained load?
No. We ran all three machines for eight consecutive minutes at full occupancy and none of them dropped below a 100% CPU speed limit at any point. The four-year-old M1 Pro finished its last round 1.0% faster than its first.
What we could not test
Our three machines are not a controlled experiment, and we are not going to present them as one. Four differences sit inside these numbers:
- The M1 Pro has 16GB of memory. The other two have 48GB. None of our tests pushed any machine into swap, so we do not believe memory drove these results, but we cannot rule it out and we did not measure memory pressure.
- The M1 Pro was running macOS 26.5.1 while the other two ran 26.6.2 when these September figures were taken, so part of that gap could have been the operating system rather than the chip. Resolved in October — see the update above, where all three ran macOS 27.0.1.
- The M5 Pro is a 16-inch machine; the other two are 14-inch. The 16-inch chassis has more cooling. This matters most for the sustained-load row, where the larger machine has an advantage that has nothing to do with its processor.
- The M4 Pro had no video encoding tools installed, so two rows of our table are blank for it rather than estimated.
We would rather publish a table with holes in it than fill the holes with guesses.
How we tested
Every machine ran the same three workloads, generated on the machine itself so no files moved between them: a 512MB text file compressed with zstd 1.5.7 at level 12, single-threaded and then across all cores; a 30-second 1080p clip encoded twice, once in software and once through the hardware encoder; and Speedometer 3.1 in Safari. Each result is the median of three runs, and we recorded the spread.
The sustained test ran the all-core compression on repeat for eight minutes and counted completed passes. We also timed the first and last pass to check for thermal throttling. None of the three machines throttled — the M5 Pro finished 0.5% faster at the end than the start, the M4 Pro was 0.5% slower, and the M1 Pro 2.1% slower. All three are within what we would call flat.
Every machine was plugged in, idle, with applications closed. That last condition turned out to matter more than we expected: our first attempt measured the M5 Pro while it was in normal use, and it came out slower than the M4 Pro — a result that is obviously wrong and that cost us a full re-test. Background applications shifted the single-core figure by 20%.
Machines: MacBook Pro 16-inch M5 Pro (48GB/1TB, battery health 100%, 5 cycles), MacBook Pro 14-inch M4 Pro (48GB/1TB, 99%, 72 cycles), MacBook Pro 14-inch M1 Pro (16GB/512GB, 86%, 264 cycles). Tested September 14, 2026.
Related measurements from the same machines: the CPU cores Apple renamed, eight minutes flat out, what 264 cycles did to a battery, and what happened when we updated all three to the 2026 release.