DeviceBenchFree memory test online
Memory test online that grows the grid with the pattern
Tiles flash across a grid, go dark, and you tap back the ones that lit — three of them on level one, one more on every level after that, and three wrong taps ends the run. Order is never checked, because the pattern arrives all at once rather than one tile after another, so the level you stop on describes how much of a layout you can hold at a glance. The grid steps up from 3×3 to 8×8 as the pattern grows and never lets more than 45% of the cells light at once; what no web page can do is tell a remembered tile from a lucky one, which is why a level you clear once is worth clearing twice.
- 100% free
- No signup
- 3 lives
- Grid grows to 8×8
- Order not scored
Level 1 · 3 tiles · 3 by 3 grid
Three tiles, then one more every level
Level 1 shows 3 tiles for 1000 ms. Tap them back in any order.
The board advances on its own only after a success; a miss holds it still so the reveal can be read. Every figure above is recomputed from this run alone, and the median recall time is the one that moves most: it drops sharply once you stop counting tiles and start seeing shapes. Cells stay square whatever the screen does, which at 8 across on a phone leaves them near 40 px — narrow enough that a thumb landing on a neighbor is a genuine cost, and the reason a wrong tap is drawn in red instead of just being subtracted from the pips.
How to take the visual memory test
The measurement is one flash long. Everything below is about not wasting it.
Take the shape, not a count
Level one lights three tiles for 1,000 ms and level ten lights twelve for 1,900 ms, which sounds like plenty until you try to count them. Counting is the trap: by the time you reach the fourth tile you have stopped looking at where the fourth tile was. Let the lit cells settle into a figure instead — a row with a corner hanging off it, a diagonal with one stray — because a figure survives the dark and a list of coordinates does not.
Tap them back in whatever order they come to you
Nothing here checks the sequence, so start with the cluster you are surest of and leave the outlier until last. A tile you only half-remember is worth tapping after the ones you know rather than before, because a wrong tap ends the level on the spot and takes the tiles you had already found with it. The counter above the grid says how many are still outstanding, which is the figure to reconcile against the shape you kept.
Read the reveal after a miss
A cleared level deals the next one by itself, but a lost level stops and redraws the pattern with your wrong tap marked in red — the only diagnostic this test offers. Two failures score identically and mean nothing alike: a wrong tile immediately beside a real one is an aiming problem, common on a phone where the cells are around 40 px, while a wrong tile on the far side of the grid means the figure you kept was not the figure you were shown.
Technical specifications
| Level 1 | 3 tiles on a 3×3 grid, lit for 1,000 ms |
|---|---|
| Growth | One tile per level; the grid steps to 4×4 at level 3, 5×5 at level 6, 6×6 at level 10, 7×7 at level 15 and stops at 8×8 |
| Maximum fill | 45% of the cells — the grid grows before any pattern would need more |
| Flash duration | 700 ms plus 100 ms per tile, ceiling 2,500 ms, so the time per tile barely moves |
| Lives | 3. A tile that never lit costs one and ends the level; the same level is then dealt again |
| Order | Not scored — the tiles appear together, so there is no sequence to get wrong |
| If the tab goes to the background | The flash is timed in animation frames, which the browser stops sending; the pattern goes dark with the tab and the level is re-dealt at no cost |
| Where the run lives | This tab's memory only — a reload puts you back on level 1 with three lives |
Frequently asked questions
Does the order I tap the tiles in matter?
No — the set is scored and the sequence is ignored. Every tile in the pattern lights at the same instant, so there is no order in what you were shown and nothing for an order in your answer to be compared against. This is the single biggest difference between this test and the block-tapping tasks it gets mistaken for, and it is why a level here is not comparable with a span from one of those.
What level do people usually reach?
Most runs end somewhere between level 8 and level 12, which is ten to fourteen tiles on a 5×5 or 6×6 grid. Below level 6 is usually a first attempt rather than a limit: the flash is over before a newcomer has decided what to look at, and a second run on the same afternoon typically adds two or three levels for nothing but knowing what to expect. Past level 16 the grid is 7 across and the patterns stop having tidy shapes at all.
Why does the grid get bigger instead of just adding tiles to a small one?
Because a crowded grid quietly turns into a different puzzle. Once a pattern covers much more than half the cells, the cheapest thing to memorize is the handful of dark ones, and a level that is really asking you to remember four gaps is not measuring what level five measured. Capping the fill at 45% and growing the grid keeps every level the same kind of problem, only larger — which is also why the tile count and the grid size are both shown above the board.
I lost a life on a tile right next to one that lit. Is that my memory?
Probably not — that is the signature of a targeting error rather than a recall error. On an 8×8 grid at phone width each cell is roughly 40 px, under the 44 px that touch guidelines ask for, so a thumb landing on the neighbor is a real and frequent failure. The reveal after a miss exists to tell the two apart: a wrong tile touching a correct one is your hand, a wrong tile several cells away is the pattern.
Is this the Corsi block test?
No, and the difference is not cosmetic. Corsi block-tapping, described by Philip Corsi in 1972, has an examiner touch blocks one after another and asks for the same order back; healthy adults manage about five. This test shows every tile simultaneously and accepts any order, which makes it a pattern-span task in the mould of the Visual Patterns Test rather than a spatial-sequence one. The two dissociate in patients — someone can be badly impaired on one and near-normal on the other — so they are measuring different machinery.
Does staring at the spot after the tiles go dark help?
For about a quarter of a second it genuinely does, and after that it stops. George Sperling showed in 1960 that a briefly flashed array leaves a near-complete visual trace that can be read out for roughly 250 to 300 ms before it decays past use. That is why the first instant after the grid goes dark feels unfairly rich compared with the second one. Blinking or looking away shortens it, so the useful habit is to keep your eyes still until you have started tapping.
Is my level saved, ranked or compared with anyone?
No — there is no account, no leaderboard and nothing written to your browser's storage, so the run exists only while this tab does. Reloading returns you to level 1 with three lives because there is nowhere for the previous run to have been kept. Copy hands the summary to your clipboard, and that is the only door out.
About pattern span, and why order is left out of it
What a simultaneous grid measures has its own name and its own literature: visual pattern span, formalized by Della Sala, Gray, Baddeley and Wilson in 1997 as the Visual Patterns Test, where a checkerboard of filled and empty cells is shown for a few seconds and reproduced from memory. Adults on the paper version typically top out around nine or ten filled cells. That is a different store from the one Corsi block-tapping loads, where blocks are touched in sequence and the average adult manages about five — and the two come apart cleanly enough in brain-injured patients that they are treated as separate components of visuospatial working memory. This test deliberately sits on the pattern side of that split, which is the whole reason it accepts your taps in any order: scoring a sequence you were never shown would be inventing a requirement.
The exposure is the setting that decides which memory is being tested, and most versions of this game leave it fixed. In the first quarter-second after the grid goes dark you are not remembering at all — you are still reading a decaying visual trace, the effect Sperling demonstrated in 1960 with partial-report arrays, and it holds far more than you can ever report before it fades. A flash of 200 ms therefore measures how fast you can read that trace, while a flash of two seconds measures deliberate encoding. Growing the exposure by 100 ms per tile keeps the ratio steady, so level twelve is a harder version of level three rather than a different task wearing its costume. One thing worth ruling out before you blame your memory: a dim or badly set panel makes a low-contrast flash genuinely harder to read, and a full white field next to a full black one will show you in a second whether your display has the contrast the test assumes.
The level you finish on is worth treating as a range rather than a figure. Retest gains of one to three levels are ordinary and mostly reflect learning the flash rather than improving anything, and the gains do not travel: getting better at grids does not make you better at names, phone numbers or where you left your keys, which is the finding that sank most of the commercial brain-training claims. If you want the same underlying question asked in a way that does score order, the chimp test hides numerals behind blocks and demands them back in sequence, and the number memory test runs the verbal version, where the limit is set by how fast you can say the digits rather than by how much of a layout you can see.
What happens to the patterns you tap
Every number on this page is worked out by JavaScript running in the tab you are reading it in. Nothing you type, paste or open is uploaded, logged or kept, which is also why the tools carry on working after you disconnect from the network.
The pattern is generated in this tab by a shuffle of the cell indices, and it is thrown away the moment the level ends. Your level, your lives and your timings are React state and nothing more — there is no account to attach them to and no leaderboard to submit them to, which is why closing the tab is the whole of the deletion process.