Memory
Chimp Test
Numerals scatter across the board and sit there as long as you want. The moment you touch the 1, the rest turn into blank squares — and now you tap them in order from memory alone. Every other version of this test tells you how many you got. This one also tells you how long you stared first, which turns out to change everything.
Chimp test
The numerals 1 to 4 appear on the board and stay put for as long as you want. The moment you touch 1, the rest go blank — and now you tap them in order from memory. Each round you clear adds another numeral, up to 9. Miss twice at the same level and you're out.
We also time how long you stare before that first touch. Look for a minute and you'll get further than someone who glanced for two seconds — so the count on its own doesn't mean much. Both numbers are your result.
What your equipment contributed
- Your display
- not measuredassuming a typical 60 Hz panel
- Study clock
- ±16.7 msstart quantised to a frame
- Board rendered at
- not measuredCSS pixels wide, 6×5
Your equipment is unusually quiet on this page, and that is worth saying plainly. The study clock starts on a frame your display can present and stops inside the touch handler, so the display adds up to 16.7 ms and your input path adds more we cannot measure — against a study period of hundreds or thousands of milliseconds, both are noise. The limit that matters here is not your hardware. It is that the exposure is yours to set. The experiment this test is named after fixed it at 210 ms for 5 numerals and masked the board automatically; this version, like every browser version, waits for you. That is why the milliseconds are printed beside the count rather than left out of the result. Board size is the second thing we can only half state: yours is whatever your viewport allows across, a CSS pixel is not a fixed physical length, and nothing in a web page can measure how far away you are sitting.
How this is measured
What starts the clock, and what stops it
The numerals are committed to the page, and the study clock is stamped inside a requestAnimationFrame callback, so the start is pinned to a frame your display can actually present rather than to an arbitrary moment between two. It stops on your touch of numeral 1 — performance.now() as the first statement in the handler, before any state change or render work, which is the same rule every timed test on this site follows. That single touch is both the end of the measurement and the event that masks the board.
Why the viewing time is half the result
The published task fixed the exposure: 5 numerals shown for 210 ms, or 430, or 650, then masked automatically. That constant is what let one subject's accuracy be compared with another's at all. Self-paced versions — every browser one, ours included — hand the exposure to the person being measured, so “9 numerals” is not one result but a family of them, ranging from a glance to a minute of study. We time the look and print it beside the count instead of quietly dropping it. Both numbers go into the share text, because a count travelling on its own is exactly the thing this page is about.
How the look is aggregated, and what is discarded
The reported figure is the median study time per numeral across your rounds — median because study times are right-skewed, and dividing by the numeral count because 5 numerals in two seconds and nine numerals in two seconds are not the same exposure. Every round with a first touch counts, including the ones you failed: the look happened either way, and using only the successful rounds would report the exposure of the rounds that went well as if it were the exposure of the run. One thing is discarded — a study period longer than 60 seconds, which measures an interruption rather than a look. Such a round still counts towards the score, and the result tells you how many were dropped and why.
The board, the ladder, and how the test ends
One fixed 6×5 board at every level, so a longer round is not also a more spread-out round — the only thing a level changes is how many numerals are on it. Levels start at 4 numerals and add one for each round completed in order. Two misses at the same level end the test, the same stopping rule as the site's other span tests: one slip is weak evidence, which is why published span procedures give more than one attempt at a length. Your score is the largest level completed. A wrong touch ends the round immediately, and the result shows how far you had got.
Why the ladder stops at 9
Ten is the first numeral that needs two digits, and a two-digit numeral in a small square is a different perceptual event from a one-digit one — it has to be read, not merely seen. A level of fifteen would still produce a number, and that number would be part memory and part legibility with no way to separate them. The published task uses the numerals 1 to 9; so do we, and reaching the top is reported as ≥9. Other versions of this test run to twenty-five and beyond. That is a decision to keep the ladder going, and it is the one we decline.
What guessing could get you
Numeral 1 is visible when you touch it, so what is really guessed is the order of the masked squares that follow — exactly (n−1)! equally likely orders, which is 1 in 6 at the opening level and 1 in 40,320 at nine. That is arithmetic, not an estimate, and it is printed with your result. The practical reading: luck can buy you the bottom of the ladder and essentially nothing above it, because a level is only reached by completing the one below.
What we cannot tell
We cannot tell whether you memorised positions or said them to yourself — “top-left, then two down the right” is a verbal strategy and it works here, which is one more reason this is not the laboratory task. We cannot tell whether you had done this before, and practice matters more on this task than on anything else on the site: see the section below. We cannot know your board's physical size or your viewing distance. And in the recall phase we deliberately do not announce the numerals to a screen reader — doing so would hand over the very thing being remembered. During the study phase the numerals are plain text on screen and are read normally.
How your result compares
Not to the chimpanzee, and not to a percentile. This is the second page on this site to show no reference range — but for a different reason from our visual memory test, where the norms exist and we haven't read them. Here we have read the study, it is short, it is clear, and it measured something this test does not.
Why the famous result is not a range for your score
Different dependent variable. Inoue and Matsuzawa held the number of numerals constant at 5 and measured accuracy at a fixed exposure. This test holds accuracy constant — you either complete a level or you don't — and measures how many numerals you can reach. There is no arithmetic that turns one into the other.
Different independent variable. Their exposure was fixed by the apparatus at 210, 430 or 650 ms and masked automatically. Yours is whatever you chose. A score obtained under a free parameter cannot be placed against results obtained by controlling it.
And the comparison was about practice, not species. Ayumu had many sessions on the task before those measurements; the nine human subjects had none. When Silberberg and Kearns gave two humans practice on the same task, their accuracy matched Ayumu's. The headline that made this test famous did not survive its first replication attempt, and any site that invites you to “beat the chimp” is selling a result that was corrected in 2009.
The one figure we will put beside yours
Their shortest condition allowed 210 ms for 5 numerals, which is 42 ms per numeral. That division is ours, not theirs, and we label it as ours everywhere it appears. It is not a threshold, not a pass mark, and not a score — the study measured accuracy at that exposure, so the figure describes the apparatus of the experiment rather than anybody's performance in it.
It is on the page for one purpose: so you can see your own viewing time on the same scale as theirs. Most people taking a self-paced version give themselves an order of magnitude more. That comparison is about exposure, never about memory, and the result panel says so in as many words.
And no percentile from our own visitors, ever. We could compute one tomorrow — self-selected, unvetted, taken on every size of screen, and with each visitor free to choose their own exposure, which is the one variable that would dominate it. It would look exactly as authoritative as the ones our competitors show. That is precisely the objection.
The experiment this test is named after: Inoue, S., & Matsuzawa, T. (2007). Working memory of numerals in chimpanzees. Current Biology, 17(23), R1004–R1005. (cited for the procedure — the masking task, the 210/430/650 ms limited holds, the 5 numerals drawn from 1–9, Ayumu's roughly 79% accuracy at 210 ms and the nine human subjects who did not approach it — and for the one exposure figure derived above. Not for a range.)
The replication that corrected the headline: Silberberg, A., & Kearns, D. (2009). Memory for the order of briefly presented numerals in humans as a function of practice. Animal Cognition, 12(2), 405–407. (cited for the finding that two humans given practice on the same task matched Ayumu's accuracy, and that the original human subjects had no practice.)
Last reviewed: July 2026
Frequently asked questions
Is this the test where a chimpanzee beat humans?
It is named after that study, and it is not that study. Inoue and Matsuzawa (2007) showed five numerals for a fixed 650, 430 or 210 ms and then masked them automatically — the young chimpanzee Ayumu stayed at roughly 79% accuracy even at 210 ms, while none of the nine human subjects came close at that duration. The fixed exposure is the experiment. Every browser version, this one included, leaves the numerals up until you touch the first, so you choose your own viewing time and the comparison is unavailable. There is a second reason the headline does not survive: Ayumu had many sessions of practice before those measurements and the humans had none, and when Silberberg and Kearns (2009) gave two humans practice on the same task, their accuracy matched Ayumu's.
Why do you time how long I look before the first touch?
Because without it the score cannot be read. Nine numerals recalled after half a second and nine numerals recalled after eleven seconds of staring are different events, and both come out of every other version of this test as the identical number "9". Since the browser version gives the exposure away, we measure it instead of pretending it away: the clock starts on the frame the numerals are presented and stops inside the handler for the touch that masks them, and your result is the pair — the count, and the median milliseconds per numeral you gave yourself. The milliseconds are on the share text too, for the same reason.
Why does this stop at 9 when other chimp tests go to twenty-five?
Because ten is the first numeral that needs two digits, and a two-digit numeral in a small square has to be read rather than merely seen — especially on a phone. Levels above nine still produce a number; that number is part memory and part legibility, with no way to separate the two. The published task uses the numerals 1 to 9, and so do we. Reaching the top is reported as "≥9" rather than "9", because at that point we have stopped measuring you and started measuring the ladder.
Could I pass a level by guessing?
At the opening level, sometimes. Numeral 1 is visible when you touch it, so what is actually guessed is the order of the remaining squares after the mask — three of them at the opening level, which is 1 in 6. It stops mattering immediately, because the number of possible orders is (n−1)! and that grows fast: 1 in 120 at six numerals, 1 in 40,320 at nine. Since a level is only reached by completing the one before it, luck can buy you the bottom of the ladder and essentially nothing above it. The odds for your own final level are printed with your result.