It depends on how hard you went and how close to bed you finished, and for most ordinary evening sessions the answer is no. Controlled studies have generally found that evening exercise does not harm sleep and may modestly improve parts of it, while a very large analysis of real-world wearable data found that strenuous sessions ending within about four hours of sleep onset are associated with worse sleep. Those two results sound contradictory and are not, which is the interesting part of this question.
If you train after work because that is when the day releases you, this is the sort of thing you have probably been given confident advice about in both directions.
What the controlled studies found
The standard reference is a 2019 systematic review and meta-analysis in Sports Medicine by Stutz and colleagues. They screened 11,717 references and pooled 23 studies of a single evening exercise session against a no-exercise control in healthy adults.
Compared with control, evening exercise increased rapid eye movement latency by about 7.7 minutes and slow-wave sleep by 1.3 percentage points, and decreased stage 1 sleep by 0.9 percentage points. Slow-wave sleep going up and light stage 1 sleep going down is a shift towards deeper sleep, not away from it. The authors’ conclusion was that the studies reviewed do not support the hypothesis that evening exercise negatively affects sleep, and that if anything the reverse holds.
They did name an exception, in the same sentence: sleep-onset latency, total sleep time and sleep efficiency might be impaired after vigorous exercise ending an hour or less before bedtime. Hold on to that, because the newer work turns it from a footnote into the main finding.
A 2021 review focused specifically on high-intensity evening exercise reached a similar conclusion, reporting that acute high-intensity exercise before bed does not disrupt subsequent night-time sleep in healthy young and middle-aged adults who are good sleepers.
What four million nights found
In 2025, a study in Nature Communications took a different approach. Rather than bringing a dozen people into a lab, it analysed data from 14,689 physically active people wearing a biometric device over a year, totalling 4,084,354 person-nights.
The finding was a dose-response relationship. Later exercise timing and higher exercise strain were associated with delayed sleep onset, shorter sleep duration, lower sleep quality, higher night-time resting heart rate and lower night-time heart rate variability. The size is not trivial. Doing maximal rather than light exercise finishing two hours before habitual sleep onset was associated with sleep onset 36 minutes later and sleep duration about 5.4 per cent shorter, roughly 22 minutes. Finishing two hours after habitual sleep onset, the gap widened to 80 minutes later onset and about 42.6 minutes less sleep.
And the clean line: regardless of strain, exercise bouts ending four or more hours before sleep onset were not associated with changes in sleep.
Before leaning on that study, one disclosure. Two of its authors are affiliated with WHOOP, the company that makes the wearable the data came from, and the paper states this as a competing interest while noting the company did not otherwise shape the design, analysis or decision to publish. The data is also proprietary and access runs through an application process. That does not make the finding wrong, and the sample size and preregistered-style rigour are real, but a commercial relationship to the measuring device is something a reader should be told.
Why both results are true
The reconciliation is in the 2025 paper itself, and it is one line: 82 per cent of the studies in the earlier meta-analyses would be categorised as light or moderate strain.
So the lab work did not test what the wearable study tested. It mostly tested moderate sessions and found them harmless, which they are. The newer study extended the range upward into genuinely hard, long sessions and found that the harm shows up there. The 2025 authors also note that when light or moderate exercise ended more than about two hours before habitual sleep onset, sleep onset and quality were similar to non-exercise nights, which is exactly what the meta-analyses said.
Two other things are worth separating out honestly. The wearable study is observational, so it shows association rather than causation, and the authors could not control for confounds like bright light exposure in late-night gyms. They also could not measure sleep onset latency directly, because the device does not detect when someone first tried to sleep. And the sample was physically active people, which the authors flag as unrepresentative. Their own prediction is that the effect would be stronger, not weaker, in less-trained people, since autonomic recovery after exercise is typically slower in those who train less often.
What this suggests in practice
The 2025 paper’s own recommendation is to finish exercise four or more hours before sleep onset where possible, and to choose lighter sessions when exercising inside that window. Its worked example: choosing light rather than maximal strain for a session ending two hours before habitual sleep onset was associated with sleep onset 36 minutes earlier, 5.4 per cent more sleep, a 6.8 per cent lower night-time resting heart rate and 14.1 per cent higher heart rate variability.
Note what is not in that advice. Nobody is telling you to stop exercising in the evening. Across both bodies of work, moderate evening sessions come out fine, and the meta-analysis found evening exercise nudged sleep architecture in a favourable direction.
If you consistently cannot fall asleep, wake unrefreshed, or suspect a sleep disorder, exercise timing is unlikely to be the whole story, and weekend catch-up sleep has limits of its own, and a doctor or sleep clinician is a better next step than rearranging your gym schedule.
Common questions
Is a hard gym session at 9pm actually a problem?
It depends on your bedtime, not the clock. A 2025 analysis of 4,084,354 person-nights from 14,689 physically active people found that later timing combined with higher strain was associated with delayed sleep onset, shorter and lower-quality sleep, higher night-time resting heart rate and lower heart rate variability, in a dose-response pattern. Maximal rather than light exercise finishing two hours before habitual sleep onset was associated with about 36 minutes later onset and 5.4 per cent less sleep. Sessions ending four or more hours before sleep onset showed no association with changes in sleep at any strain. Two caveats: the study is observational, so it shows association rather than cause, and two of its authors are affiliated with the company that makes the wearable the data came from.
Why do older articles say evening exercise is fine?
Because for the intensities they tested, it is. A 2019 meta-analysis of 23 studies found evening exercise increased slow-wave sleep by 1.3 percentage points and reduced light stage 1 sleep by 0.9 percentage points compared with no exercise, and concluded the evidence does not support the idea that evening exercise harms sleep. The 2025 wearable study points out that 82 per cent of the studies in those meta-analyses would count as light or moderate strain. So the older conclusion is sound for moderate sessions and simply does not cover the hard, long ones. The 2019 review did flag that sleep onset, total sleep time and sleep efficiency might be impaired after vigorous exercise ending an hour or less before bed.
Does it matter what kind of exercise?
Less than how hard and how late, on the current evidence. The 2019 meta-analysis found running was associated with less wake after sleep onset than cycling, but reported that all of its significant moderating effects disappeared when a single study was removed, which is a strong signal not to build anything on them. The 2025 study measured strain from heart rate rather than exercise type, and its authors note that this method may underestimate the strain of strength training involving very brief bursts of high effort, so weights specifically are less well characterised than cardio. Our own reading rather than a study finding: the strain framing is more useful here than the sport, because it is what both papers converge on.
What if I have no choice but to train late?
The 2025 paper’s suggestion is to pick lighter sessions inside the four-hour window rather than skip exercising. In its data, choosing light instead of maximal strain for a session ending two hours before habitual sleep onset was associated with sleep onset 36 minutes earlier, 5.4 per cent more sleep and 14.1 per cent higher night-time heart rate variability. The paper is observational, so these are associations, and two authors are affiliated with the wearable company whose data was used. It also notes that late-evening training often comes bundled with other things that affect sleep, including bright gym lighting and caffeine or carbohydrate intake around the session, which it could not control for. Persistent sleep problems are worth raising with a clinician rather than solving by schedule alone.
The variable is strain, not the hour
The question is usually asked as though evening were the problem. Both bodies of evidence say it is not. What the data actually tracks is how much physiological load you put on your body and how much time you left it to come down before asking it to sleep.
That is a more useful way to hold it, because it is adjustable in more than one direction. You can move the session earlier. You can make it lighter. On the nights you can do neither, you can know what you are trading, which is roughly half an hour of sleep rather than something dramatic.
The version of this advice that tells you not to exercise at night has the wrong variable in the wrong place. Exercising in the evening is not the risk. Going maximal an hour before bed, repeatedly, is.
Leave a Reply