Self Lab

Evidence-based self-improvement. No mechanism, no article.

Esc or click outside to close

Sleep & Energy · Mechanism

Caffeine Half-Life: Why Your Cutoff Is a Guess

Ten men, the same caffeine dose, half-lives from 2.7 to 9.9 hours. Why the 6-hour cutoff rule oversells timing, and what the dose data actually shows.

Alex by Alex July 29, 2026 Updated August 4, 2026 18 min read

Ten healthy young men. One lab. The same dose of caffeine, measured to their body weight. Their bodies cleared it at half-lives ranging from 2.7 to 9.9 hours (Blanchard & Sawers, 1983).

Same dose, nearly a fourfold spread. That’s the crack in the number everyone repeats.

“Caffeine’s half-life is 5 hours” gets handed to you like a personal fact. It’s a population midpoint. And the famous rule built on top of it, no coffee within 6 hours of bed, comes from a 12-person study using a dose most people never drink in one go.

This guide names the mechanism, then shows what the better-powered trials found. Like most habit advice, the popular version skips the part that actually matters: the size of the dose, not the hour on the clock.

A quick note: this is education, not medical advice. If you suspect a sleep disorder, that’s a conversation for a clinician, not a blog.

The Bottom Line

  • Caffeine doesn’t add energy. It blocks the tiredness signal that’s already piling up.
  • The 5-hour half-life is an average. A 1983 pharmacokinetic study found 2.7 to 9.9 hours across ten men on one dose (Blanchard & Sawers, 1983).
  • Dose does most of the work timing gets credit for: 100 mg four hours before bed changed nothing measurable in a 2025 trial.
  • 400 mg at the same hour cost 50.6 minutes of sleep in that same trial.
  • Deep sleep can drop while your total hours look fine. Judge by mornings, not by how fast you fell asleep.

In this guide:

What does caffeine actually do to sleep?

Caffeine doesn’t hand you energy. It hides the fact that you’re tired. Every hour you’re awake, a molecule called adenosine builds up in your brain and pushes on the sleep switch. Caffeine parks in adenosine’s spot and blocks that push. Researchers call it an adenosine antagonist that acutely reduces sleep pressure (Gardiner et al., 2023, Sleep Medicine Reviews).

Here’s the part that matters for your night. Blocking the signal doesn’t cancel the debt. The tiredness kept accumulating the whole time. When the caffeine clears, the bill is still sitting there, which is why the 4pm coffee often ends in a 3am reckoning.

So a “caffeine crash” isn’t caffeine leaving. It’s you finally feeling what was true two hours ago.

That mechanism explains why the timing question is even a question. If caffeine simply added alertness, clearing it would be neutral. Because it masks a signal instead, the thing you care about is how much of the blocker is still on board when your head hits the pillow. Which brings us straight to half-life.

Is the 5-hour half-life your number?

Probably not. Half-life means the time it takes your body to clear half of what you took. In adults it’s about 5 hours (StatPearls, updated 2024). That figure is a midpoint drawn across a wide, messy population, and it gets quoted as though it were printed on your birth certificate.

Run the textbook arithmetic and you get a tidy story. A 200 mg coffee at 4pm leaves roughly 100 mg at 9pm and roughly 50 mg at 2am. Treat that as arithmetic, not as a measurement of you. In a 1983 pharmacokinetic study, ten healthy young men on an identical 5 mg/kg dose cleared it at half-lives from 2.7 to 9.9 hours (Blanchard & Sawers, 1983).

On the fast end, that coffee is basically gone by 10pm. On the slow end, most of it is still with you at bedtime. Same cup, same person’s body weight, completely different night. And a rough night is where the morning argument starts, though the snooze button is not the villain there either.

What stretches or shrinks it?

Four things move the number reliably, and none of them are on the chart you saw on Pinterest.

What you’re working withWhat happens to caffeine clearanceSource
SmokingHalf-life cut by up to 50% versus nonsmokersStatPearls (2024)
Low-dose combined oral contraceptivesMean half-life 5.37 h → 7.88 h in 9 users vs 9 matched controlsAbernethy & Todd (1985)
Third-trimester pregnancyHalf-life up to about 15 hoursStatPearls (2024)
Liver disease, or drugs that block CYP1A2Clearance slows, half-life stretchesStatPearls (2024)

CYP1A2 is just the liver enzyme that does most of the caffeine breakdown. The contraceptive finding is real but narrow: 18 people, 1985, older low-oestrogen formulations. Don’t stretch it to cover every modern pill. The pregnancy figure is pharmacology, not advice.

Can a DNA test tell you your curfew?

No. Genetics is a genuine moderator, and it’s been oversold hard. A 2007 study found that people who called themselves caffeine-sensitive showed worse sleep with habitual intake, while self-described insensitive people didn’t, and that the two groups differed in their ADORA2A gene variants (Rétey et al., 2007).

That’s a group-level pattern, not a personal readout. The 2025 dose trial below genotyped its participants and was explicitly underpowered to say anything about genetic effects. No consumer test currently tells you your hour.

Where did the 6-hour cutoff rule come from?

From one small trial in 2013, and it’s more interesting than the headline. Researchers gave 12 healthy sleepers 400 mg of caffeine in pill form at 0, 3, or 6 hours before their usual bedtime, plus a placebo night (Drake et al., 2013, Journal of Clinical Sleep Medicine). Measured sleep time dropped by roughly 66 to 72 minutes.

Note what 400 mg actually is: about four 250 mL cups at the 107 mg reference, or one large commercial 16 oz brew, swallowed at once. Note also the instrument. Sleep was recorded at home with a validated wireless dry-EEG headband, not full in-lab polysomnography.

Now the odd bit nobody quotes. The loss was about the same at 0, 3 and 6 hours. If timing were the engine, that line should slope. The 2023 meta-analysis reviewing this trial found no clear timing relationship inside it.

So the “6 hours” in the famous rule isn’t the point where damage starts. It’s simply the longest gap the researchers happened to test, and even there the sleep loss was still showing up.

One more detail worth sitting with. The participants’ own sleep diaries caught the disruption at 0 and 3 hours. At 6 hours, the self-reported effect didn’t reach significance (p=0.08) while the objective loss did. They lost the sleep and largely didn’t notice.

The authors listed their own limits plainly: 12 people, no blood draws, so nobody’s actual metabolism was measured.

How much sleep does caffeine actually cost?

The best pooled estimate is about 45 minutes. A 2023 systematic review in Sleep Medicine Reviews combined 24 controlled crossover studies in healthy adults aged 18 to 65, with 16 of the 24 rated low risk of bias (Gardiner et al., 2023). Total sleep time fell 45.3 minutes on average. Sleep efficiency dropped 7.0%.

What a caffeine night looks like, pooled Pooled effects versus placebo across 24 controlled crossover studies in Gardiner et al. 2023: total sleep time minus 45.3 minutes, deep sleep minus 11.4 minutes, sleep onset latency plus 9.1 minutes, wake after sleep onset plus 11.8 minutes, and no significant effect on REM sleep duration or proportion. What a caffeine night looks like Average change vs placebo, 24 controlled crossover studies Total sleep 45.3 min less Deep sleep 11.4 min less Falling asleep 9.1 min longer Awake at night 11.8 min more REM sleep no significant change less sleep 0 more Source: Gardiner et al. (2023), Sleep Medicine Reviews
Deep sleep takes the hit. REM largely doesn't. Source: Gardiner et al. (2023).

Kill one myth right here. Caffeine is widely blamed for wrecking REM sleep, and the pooled data doesn’t support it: no significant effect on REM duration or proportion, and REM onset latency moved by a non-significant 1.5 minutes. What did move was deep sleep, down 11.4 minutes, while light stage-1 sleep rose 6.1 minutes. Your night gets shallower, not dreamless.

The review also modelled dose and timing. Every extra hour between your last dose and bed shaved 2.8 minutes off the sleep loss. Every extra milligram added 0.2 minutes.

Then came the number that went viral: cut coffee 8.8 hours before bed, so roughly 1pm. Read how it was built. The cut-off is the first timepoint where the confidence interval around the modelled effect stops crossing zero, assuming a 10pm bedtime and a 107 mg cup. Same model, 217.5 mg of pre-workout: 8:50am. Same model, a 47 mg cup of black tea: no identifiable cut-off at all.

That last result is the tell. A cut-off that vanishes when you shrink the cup was never really about the clock.

Is it the timing or the dose?

Dose. In 2025, the same lead author ran the trial the internet skipped: 23 healthy men, at home, with partial polysomnography, across seven conditions of placebo plus 100 mg or 400 mg at 12, 8 and 4 hours before bed (Gardiner et al., 2025, Sleep). At 100 mg, nothing reached significance. Not one measure, not at any timing.

Dose does the work the clock gets credit for Minutes of total sleep lost versus placebo in Gardiner et al. 2025: 400 mg taken four hours before bed reduced total sleep time by 50.6 minutes, a significant effect. At eight hours the reduction was 28.7 minutes and at twelve hours 30.0 minutes, neither statistically significant for total sleep time, though other measures were still affected: sleep efficiency fell 6.9% and night-time waking rose 29.1 minutes at eight hours, and latency to persistent sleep rose 15.3 minutes at twelve hours. 100 mg produced no significant effect on any measure at any timing. Dose beats the clock Total sleep lost vs placebo, 23 adults, at-home sleep recording 400 mg, 4 h before bed 50.6 min lost 400 mg, 8 h before bed 28.7 min, not significant 400 mg, 12 h before bed 30.0 min, not significant 100 mg, any timing no measurable effect 0 30 min 60 min Source: Gardiner et al. (2025), Sleep 48(4)
Same trial, two doses. The small one is invisible to the measurement. Source: Gardiner et al. (2025).

At 400 mg the picture changed. Four hours before bed cost 50.6 minutes of sleep, pushed sleep efficiency down 9.5%, added 14.2 minutes to falling asleep and 26.2 minutes of night-time waking. Perceived sleep quality dropped 34% in that condition alone. At 8 hours the sleep-time loss (28.7 minutes) and at 12 hours (30.0 minutes) missed significance, but fragmentation did not: 8 hours out, sleep efficiency still fell 6.9% and night-time waking rose 29.1 minutes (both p=0.001), and 12 hours out it took 15.3 minutes longer to reach persistent sleep.

One thing did travel across all three timings: deep sleep. On 400 mg it fell 29.7 minutes at 4 hours, 15.3 at 8 hours and 20.6 at 12 hours. A morning coffee that leaves your total hours untouched can still flatten the deepest part of the night.

The authors addressed the 8.8-hour rule themselves. Because that cut-off came from statistical modelling, they noted, the actual size of the sleep reduction “could not be determined to evaluate the clinical significance”. Translation: a boundary on a model isn’t the same as minutes lost in a bed.

I ran a scrappy version of this on myself for a month. Moving my last coffee from 4pm to 1pm did nothing I could feel. Halving it, from a double to a single, showed up within four days as fewer 3am wake-ups. That’s an n of 1 and worth exactly what you paid for it. But it matches where the controlled evidence points.

Why do some people sleep fine after coffee?

Sometimes because they genuinely do, and sometimes because they’re bad witnesses. Both are documented, and honesty means holding them together.

Start with the strongest “it’s fine” evidence. Twenty habitual consumers averaging 478 mg a day took 150 mg three times daily for 10 days, with the last dose 8 hours before bed, monitored by lab EEG (Weibel et al., 2021, Scientific Reports). Against placebo: no significant difference in total sleep time (placebo 441 minutes vs caffeine 451 minutes), sleep efficiency, deep sleep percentage or subjective quality.

Then the field data. A study of 785 adults wearing wrist actigraphy across 5,164 person-days found that caffeine within 4 hours of bedtime moved sleep duration by 1.15 minutes (Spadola et al., 2019, Sleep). Nicotine and alcohol showed real effects that night. Caffeine didn’t.

Be careful with that one. Exposure was a yes-or-no diary question, with no dose, no sensitivity and no tolerance measured, in an older group of habitual drinkers. That’s a null result from a blunt instrument, not proof of no effect.

So who’s right?

Both, in their own lane. Lab-sized single doses reliably degrade sleep. Ordinary sipped intake in habituated people often doesn’t show up in the measurements at all.

What doesn’t survive scrutiny is “I sleep fine, so I’m immune”. Drake’s participants lost over an hour of measured sleep at the 6-hour timing and their diaries barely registered it. Your self-report is the weakest instrument in the room.

Two more things you can’t lean on. Tolerance isn’t established protection: evidence for receptor adaptation comes largely from animal work, with the 2023 review noting greater uncertainty about caffeine tolerance in humans. And deep sleep is the measure most likely to shift while your total hours look perfectly normal.

What else does evening caffeine change?

It can move your body clock, not just your sleep. In a roughly 49-day laboratory protocol, a dose equal to a double espresso taken 3 hours before bed delayed the evening melatonin rise by about 40 minutes compared with placebo (Burke et al., 2015, Science Translational Medicine). In five people. Say that part out loud every time.

Five participants is tiny, and the mechanism was backed up in cells, where caffeine lengthened the internal clock’s period through adenosine receptors. For comparison, bright light in the same study delayed the clock by about 85 minutes. Caffeine is a weaker lever than light, but it points the same direction: later.

Caffeine also changes the texture of sleep. In a 2025 double-blind crossover, 40 adults aged 20 to 58 took 200 mg before two recorded nights (Thölke et al., 2025, Communications Biology). Their EEG showed more complex, less settled brain activity during sleep, mostly in non-REM. The REM effects reached significance only in the younger group.

What that means for how rested you feel is genuinely unsettled. The researchers don’t claim it, so neither will I. The practical read is narrower and still useful: the number of hours you slept is a crude summary of the night, and caffeine can change the parts that number can’t see.

How to find your own caffeine curfew

Stop counting hours and start counting milligrams. The evidence says the size of your last dose predicts the damage better than the hour you drank it. So run the experiment in that order: shrink first, move later.

Here’s what a dose actually looks like, using the reference amounts from the 2023 review.

DrinkCaffeineModelled cut-off before a 10pm bedtime
Black tea, 250 mL47 mgNone identifiable
Coffee, 250 mL107 mg8.8 hours (about 1pm)
Pre-workout serving217.5 mg13.2 hours (about 8:50am)

Those cut-offs are modelled boundaries, not physiology. Use the middle column, not the right one.

Week 1: measure, change nothing. Write down what you drink and when. Milligrams and clock time, one line each. Most people are shocked by the total, not the timing.

Week 2: halve the last dose. Keep the hour identical. A double becomes a single. In the 2025 trial, 100 mg four hours before bed did nothing measurable, while 400 mg at that hour cost 50.6 minutes.

Week 3: move it earlier, one hour at a time. Only if week 2 didn’t help. Each extra hour bought about 2.8 minutes in the pooled model, so expect small gains, not miracles.

Judge by mornings. Not by how fast you dropped off. Deep sleep is what caffeine hits hardest, and you fall asleep fine while losing it.

To make the swap stick, pin it to something already automatic. Attach it to a cue you already have: after the 3pm meeting, the kettle gets tea instead of coffee. And if your evenings die to your phone rather than your mug, that’s a different loop with a different fix.

FAQ

How many hours before bed should I stop drinking caffeine?

There’s no single hour, and the honest answer depends on dose. In a 2025 crossover trial, 100 mg four hours before bed produced no significant effect on any sleep measure, while 400 mg at that same hour cut total sleep by 50.6 minutes (Gardiner et al., 2025). Shrink the cup before you move the clock.

Does caffeine ruin REM sleep?

The pooled evidence says no. Across 24 controlled crossover studies, caffeine showed no significant effect on REM duration or proportion, and REM onset latency shifted by a non-significant 1.5 minutes (Gardiner et al., 2023). Deep sleep is the casualty instead, down 11.4 minutes on average.

I sleep fine after evening coffee. Am I immune?

Possibly, and possibly not measuring it. A 785-person field study found no association between caffeine within 4 hours of bed and that night’s sleep (Spadola et al., 2019). But in the 2013 trial, people lost over an hour of measured sleep at the 6-hour timing without reliably noticing.

Does building tolerance protect my sleep?

Not proven. The 2023 review states that the evidence for adenosine-receptor adaptation comes largely from animal studies, noting greater uncertainty about caffeine tolerance in humans. Habitual drinkers in a 10-day trial did show no change in sleep structure with the last dose 8 hours out (Weibel et al., 2021). Suggestive, not settled.

Is tea a safer evening choice than coffee?

By dose, yes. In the 2023 model, a 250 mL cup of black tea carries about 47 mg versus 107 mg for coffee, and no clear cut-off time could be identified for tea at all (Gardiner et al., 2023). That’s a statement about milligrams, not magic leaves.

The Bottom Line

The half-life chart isn’t a schedule. It’s a population average being sold to you as a personal deadline, and a 1983 pharmacokinetic study measured a spread of 2.7 to 9.9 hours in ten men on one dose. The famous 6-hour rule came from twelve people swallowing 400 mg at once. When better-powered trials separated dose from timing, dose won: 100 mg four hours before bed was invisible to the instruments, while 400 mg at the same hour took nearly an hour of sleep.

So stop negotiating with the clock and start negotiating with the cup. And watch your mornings, because the part of sleep caffeine hits hardest is the part you can’t feel it taking. Small change, run for a week, then judge.

One small action today: write down one line, the milligrams in your last caffeine of the day and the time you drank it. Just that. Tomorrow you’ll know which number you’re actually adjusting. Need a smaller starting point? Pick a two-minute version instead.

Alex is the voice of Self Lab: practical psychology for people who are done with motivational fluff.

Sources