Nine measurements · about a minute each
How fast is your brain?
Thirty seconds gets you a number. Beating it is what eats the evening.
▶ Test my reaction time measured honestly — see how
The tests
- Reaction TimeClick the moment the panel turns green. Five rounds, median scored.
- Aim TestThirty targets, one at a time. Average time and accuracy.
- Number MemoryA number flashes, then you type it back. One digit longer each round.
- Sequence MemorySquares light up in order. Repeat the pattern as it grows.
- Visual MemorySquares flash all at once. Find every one, in any order.
- Chimp TestNumbers flash, then hide. Click them in order.
- Verbal MemoryWords appear one at a time. Call each one SEEN or NEW.
- Color MatchOne tile is a slightly different shade. The gap shrinks every level.
- Typing SpeedCopy the sentence for 30 seconds. Only correct letters count.
▶ Full benchmark The eight brain tests in one sitting, scores collected into a single summary.
How this site works out a percentile
A percentile is only worth anything next to people who were measured the same way you were. That is the problem with borrowing a number from a laboratory study: the study used laboratory apparatus, and you are using a browser, a monitor and a mouse of your own. So this site is building its own reference instead. When you finish a run, one anonymous record goes out with it: which test, which difficulty, the score, whether you played on a touch screen or a mouse, the language, the date. No account, no identifier, nothing that points back at you. That pile of runs, all measured by the same clock and scored by the same rule, is the comparison this site trusts.
It stays switched off until it is honest. A percentile drawn from a few dozen runs is noise wearing a decimal point, so each test keeps its own counter and shows nothing until that test has collected enough runs to place you properly. We would rather show you a score with no percentile for a while than a confident number that moves every time somebody new plays.
Until a test reaches that point, the rule is simple. Where a published study genuinely matches how the test is played here, the percentile comes from the study and the study is linked: sequence memory sits on a sample of 340 adults, named on the methodology page. Reaction time has a famous sample of 1,469 adults, and it still shows no percentile, because browser and device delay varies too much between setups for one correction to be fair to any individual; that study stays on the page as reference for reading your own score. The rest show your score and your own trend, because their published norms describe a different task: aim depends on your pointing device, color discrimination depends on your screen, the classic digit span is read aloud by an examiner rather than typed into a browser, and the norms closest to visual memory come from a sequential task with physical blocks.
Guides
- How accurate is a browser reaction time test?A browser adds roughly 80 ms nobody spent. Where it comes from, and what we measured building this.
- How to improve your reaction timeSleep, time of day, warm-up and hardware do most of the work. Supplements do not.
- Do brain training games actually work?They make you better at themselves. Everything past that is contested.
- How we screen leaderboard nicknamesA slur got through our filter. What replaced it, and why over-blocking costs more.
Every one of them is linked to its sources, or measured by us and labelled as such. See all guides.
How to read what you get
Every result here comes with the thing most test sites leave out: where the comparison came from. Sequence memory is placed on a sample of 340 adults from a peer-reviewed paper that is linked on the page itself rather than summarised and hidden. Reaction time cites a 1,469-adult laboratory study as reference but wears no percentile, because device delay would make that number dishonest. The other seven tests show your score with no percentile attached, because no published distribution matches how they are played in a browser and we would rather show less than invent more.
The absolute number matters less than you would think. A browser adds roughly 80 milliseconds nobody spent, a 60 Hz screen costs about 8 more than a 144 Hz one, and your own trials scatter by around 40 milliseconds run to run. That is why reaction time scores the median of five rounds rather than your best click, why its early clicks are voided, and why timing refuses to run at all while the tab is in the background. The comparison that survives all of this is you against your own history on the same device. That is exactly what the score panel shows you.
What we measured while building this
The figures below come from building this site rather than from a research paper. They are listed here because a test that does not explain its measurement method cannot be verified by anyone, yet most test sites do not disclose theirs.
Timing starts the moment the pointer goes down, not on the click event that fires after release. A reaction score is the median of five rounds rather than the average. These two choices alone change the numbers. Sites that time click events and average every round will produce different values for the same person on the same hardware, which is why comparing raw millisecond scores between sites is largely meaningless.
Clicks registered within 100 milliseconds of the green signal are marked invalid, matching the threshold used to call false starts in track events. Anything above that line counts, including fast reactions or lucky guesses. A machine cannot distinguish a fast reaction from an early guess, so this test only filters out values that are physically impossible.
The wait time before the signal appears is chosen randomly between 1.5 and 7 seconds. A narrow delay window allows people to time their taps to a rhythm, yielding scores that do not reflect actual skill. Widening the delay range was more reliable than trying to detect rhythm-based tapping after the fact.
Hardware delays are not subtracted from your score. Reported delays caused by browsers and devices range roughly from 71 to 87 milliseconds and vary by setup. Applying a single flat correction would create a false level of precision. The research figures remain on the methodology page for reference.
Browsers slow down timers in background tabs enough to ruin measurements. When testing with a hidden tab, rounds that normally take 2 seconds exceeded the 25-second limit. Every timed test here refuses to start while the tab is hidden and notifies you on screen, rather than quietly saving inaccurate numbers.
On a 320-pixel screen, a 6x6 grid cannot fit 44-pixel targets for every square. Instead, square sizes are calculated from the measured width of the panel. In an earlier version with fixed sizes, squares overlapped neighboring tiles by 3 pixels, causing taps near an edge to trigger the wrong tile. That bug is why grid size calculations are now centralized in a single file.
How the leaderboards work
Each test includes a leaderboard, and submitting your score is optional. After completing a round, you simply enter a nickname. There is no account sign-up. Skipping the leaderboard costs you nothing, as your personal records are stored in your browser regardless.
Scores accumulate over time instead of resetting daily. We initially cleared scores every night, but that left the leaderboards empty and made the experience uninteresting. Once a game reaches over 1,000 participants, we will switch to a regular seasonal reset, with advance notice provided before a season ends. Hard mode uses a separate leaderboard from standard mode, as combining them would compare players completing different tasks. Typing tests are separated by language for the same reason.
Submissions are verified before they are saved. Each request includes a signature token issued when the page loaded, and scores arriving faster than physically possible for that game are rejected. We also limit multiple submissions sent from the same address within a short timeframe. These measures do not make a public leaderboard completely cheat-proof, but we prefer to state that limitation openly rather than hide it.
Nicknames are reviewed as well. Profanity and abusive names are blocked from the leaderboard. Anyone who finishes a test sees the leaderboard whether they wanted to or not, so what stays on it is our responsibility rather than theirs.