How Long People Really Last Without Blinking

Live data · last recomputed 2026-09-04

Every article about staring contests repeats the same handful of borrowed numbers: humans blink 15 to 20 times a minute, a blink takes 300 to 400 milliseconds, the world record is an hour and five minutes. All true, all copied from the same few sources, and none of it answers the question people actually ask before they try.

How long will I last?

This page answers that from the only place an honest answer can come from: the runs themselves. Every attempt on this site is judged by a face-landmark model watching the eyelid, with a threshold calibrated to that player's own eyes, and clocked by the server rather than by the player. That gives a dataset nobody else has — not because it is clever, but because we are the ones holding it.

What a typical run looks like

The median attempt lasts 9.74 seconds. Half of everyone who tries does worse than that, half does better.

The mean is 22.67 seconds, noticeably higher, and the gap between the two is the interesting part. A handful of stubborn people run several minutes and drag the average up with them. That is why "the average person lasts X seconds" is a misleading way to describe this: the average describes almost nobody. The median describes the person reading this.

The middle half of all runs — the 25th to the 75th percentile — falls between 4.47 and 21.4 seconds. If you land in that band, you are an entirely ordinary human with entirely ordinary tear film.

The whole distribution

Grouping every recorded run by how long it lasted shows the shape of the thing far better than any single number:

The cliff in the first minute is the blink reflex doing its job. Your tear film starts breaking down within seconds of the last blink, the lipid layer thins, the cornea begins to dry, and the urge arrives long before any danger does. 78.9% of runs are over inside thirty seconds. 89.5% are over inside a minute.

What the tail shows is that the reflex is suppressible, not absolute. People do get past it — just far fewer than the confident numbers floating around the internet suggest. If you have read that "most people can manage two or three minutes," this table is the correction.

Why these numbers are lower than you expected

Two reasons, and both matter if you want to compare this data to anything else.

The referee is strict. A human judge in a playground contest misses fast blinks — a 300-millisecond eyelid is genuinely hard to catch, and the person judging is usually also playing. The model here samples every frame and stops on two consecutive closed ones. Runs measured this way are shorter than runs measured by a friend, and they are shorter because they are more accurate, not because the players are worse.

These are casual attempts. Nobody on this board is preparing. There are no eye drops, no humidity control, no training block, no attempt at the conditions that produce record claims. This is what happens when an ordinary person on an ordinary afternoon decides to try. That is precisely what makes it useful: it is the distribution you belong to.

For the record attempts at the other end of the scale — and which of them survive scrutiny — see the fact-checked record ledger.

What counts as an attempt

One decision shapes every number above, so it is stated here rather than buried: a run has to last at least one second to be counted.

Without that floor, this page would have told you the median person lasts about five seconds — and that would have been a fact about our detector, not about human eyes. 108 runs, or 25.5% of everything the server has ever recorded, sit below the line. 68 of them ended inside 250 milliseconds, which is shorter than a single complete blink. Nobody loses a staring contest faster than they can blink. Those are false starts: the round begins, the player has not settled into frame yet, and the model calls it.

The cut was drawn from the shape of the data rather than picked to look good. Runs cluster hard in the first quarter-second, thin out sharply, and by one second the real distribution has taken over. One second is also more than double the length of a complete blink, so nothing below it can be an attempt at suppressing one.

The false starts are still recorded — they are how we would notice a detector drifting — they are simply not counted as attempts. Every figure on this page is computed after that exclusion, and the shortest run that does count is 1.11 seconds.

How much of this data is verified

Of 304 runs shown here, 0 were timed end-to-end by the server: the browser requested a signed clock when the run started, and the submitted duration could not exceed the time that actually passed. Those are as solid as a web leaderboard gets.

The rest predate that mechanism or were submitted without a valid token. They are kept, because throwing away real people's runs to tidy a statistic is its own kind of dishonesty — but they are capped at five minutes, which is why no thirty-four-minute fantasy appears anywhere on this page. More on how that works, and on the forged rows that made it necessary, on the about page.

This week

In the last seven days, 11 runs were recorded, and the best of them stands at 104.02 seconds. That is the number the leaderboard on the home page is asking you to beat — a real one, posted this week, by somebody with the same eyes as you.

Using this data

The aggregate figures on this page are free to cite and reuse with attribution to dontblink.fun. The machine-readable version lives at /api/stats and returns the same numbers as JSON. It contains no personal data — durations and counts only. Anything cited from here should carry the one-second floor with it, since the median moves a long way without it.

If you are a journalist, a researcher, or writing about blink physiology and want a cut of the data we do not publish here, ask: yohann@kipdev.io.

Frequently asked questions