What's a Good Memory Score? Realistic Benchmarks for All Six Tests

You have just finished a round on one of the six tests here — the Chimp Test, Sequence Memory, Number Memory, Visual Memory or Verbal Memory, or the N-Back Test — and you want to know one thing: is that number good? The answer depends on which test you played, because each one loads a different part of working memory and scales its difficulty differently.

Two different kinds of number follow. The first three sections give the editorial tier bands the site shows after a round. The Visual Memory and Verbal Memory sections give percentiles from a published score distribution, so they are firmer. The N-Back section gives no percentile at all, and says why.

Chimp Test: tile count

The Chimp Test starts at 4 tiles and adds one more each round, so your score is simply the highest count you cleared before a wrong click ended the run. In practice:

0–3 tiles — didn't survive the opening round. Completely normal on a first attempt; the layout is unfamiliar and the tiles vanish the instant you start clicking.
4–5 tiles — average. Most first-time players land here once they understand the rules.
6–7 tiles — sharp. You're reliably encoding spatial position for six or seven vanished targets, which is a solid working-memory result.
8–9 tiles — excellent, and where most regular players eventually plateau with practice.
10–11 tiles — exceptional, genuinely rare territory for a human being tested cold.
12+ tiles — the name of the game. This is the range that starts to approach what young chimpanzees demonstrated in the research the test is named after (see our history of the Chimp Memory Test).

Sequence Memory: sequence length

Sequence Memory starts with a single flashing tile and appends one more to the pattern each time you reproduce it correctly, so your score is the longest sequence you got completely right, in order.

0–4 tiles — just warming up. This covers a lot of first attempts, since remembering an ordered sequence is a different skill from remembering static positions.
5–7 tiles — average recall, and a comfortable target for casual play.
8–10 tiles — sharp memory, requiring genuine chunking strategy rather than brute-force holding of every tile individually.
11–13 tiles — excellent, where most dedicated players settle after a few sessions of practice.
14–17 tiles — exceptional, well above where most people's sequential recall tops out.
18+ tiles — grandmaster-level recall, and a genuinely impressive feat of sustained sequential memory.

Number Memory: digit count

Number Memory starts at 3 digits and adds one more each round, testing pure digit-span recall with no spatial component at all.

0–5 digits — just warming up, though 5 digits is already a completely ordinary result — this is close to a classic phone-number chunk.
6–7 digits — average recall, and squarely inside the range most cognitive research treats as typical adult digit span.
8–9 digits — sharp memory, at or above the upper edge of what most people can hold without a deliberate strategy.
10–12 digits — excellent, and where practiced players tend to plateau.
13–15 digits — exceptional, requiring real chunking technique (see our guide to improving working memory) to sustain.
16+ digits — savant-level recall, far beyond ordinary unaided digit span.

Visual Memory: the level you reach

Visual Memory flashes a set of tiles on a square grid, and you click exactly those tiles once the board resets. Level 1 flashes 3 tiles on a 3×3 grid. Each level adds one more tile, the grid gains a cell every three levels, and it stops growing at 7×7. You have three lives, and your score is the level you reached.

This test is the one place on the site where the benchmark is not editorial. The site compares your level against a model fitted to the visual-memory score histogram that Human Benchmark publishes (N = 29,243, median level 11.1). Both sites count the level reached, so no conversion is needed and the percentiles below are direct:

Level 8 — about the 7th percentile of that published distribution.
Level 9 — about the 18th percentile.
Level 10 — about the 32nd percentile.
Level 11 — the median. Half of the published scores fall below it.
Level 13 — about the 76th percentile.
Level 15 — about the 91st percentile.
Level 16 and above — the top 5 percent of that distribution.

Read that ladder rather than the rating text on the results panel. The Visual Memory tier labels are editorial, they were written before the published model was added, and they are far harsher than the data: the panel calls level 11 “exceptional” when the published median sits at exactly that level. The percentile line under your score is the number to trust.

Verbal Memory: words survived

Verbal Memory shows one word at a time, and you mark each word SEEN or NEW. Words are drawn from a pool and repeat often, so the list you have to hold keeps growing. A wrong call costs one of three lives, and your score is how many words you get through.

This test also has a published distribution behind it — the Human Benchmark verbal-memory histogram (N = 26,860, median 41.9 words) — and both sites count words the same way:

25 words — about the 21st percentile.
30 words — about the 30th percentile.
42 words — the median.
65 words — about the 75th percentile.
100 words — about the 91st percentile.
120 words and above — the top 5 percent.

One honest limit: the published histogram has a heavy first bin, because roughly 12 percent of recorded scores are under 10 words. A smooth curve cannot follow that step, so below about 25 words the model reads a few points high. Treat any percentile in the low twenties and under as approximate. Above 25 words the fit tracks the histogram closely.

N-Back: no benchmark, and why

The N-Back Test has no percentile here, and this page will not invent one. No published distribution of n-back level exists that this site could cite, so any percentile for it would be a number with nothing behind it. The other five tests get percentiles because a large public score histogram exists for each of them. N-back gets none.

What you can read instead is what the test actually measures. A session is three blocks of about twenty scored trials, and roughly 30 percent of the scored trials are targets. Because most trials are not targets, percent correct is a misleading measure: press nothing at all and you still score about 70 percent. The results panel therefore reports hits, misses, false alarms and correct rejections separately, and derives d-prime from the hit and false-alarm rates. A d-prime of 0 means no discrimination whatsoever between a target and a non-target, whatever the percentage says.

The rung moves on its own. Three clean blocks in a row — every target caught, no false alarms — moves you up one. Two missed targets inside a single block moves you down one. That ladder is the benchmark: it parks the task just past what you can comfortably do, and it keeps doing that as you improve. Most people who practise sit at 2-back or 3-back. Treat a stable 3-back with a d-prime that does not collapse in the third block as a genuinely good result, and compare it with your own past sessions rather than with a percentile that does not exist.

Where these numbers come from, and where they don't

The Chimp, Sequence and Number tiers above are not drawn from a peer-reviewed study of this specific game — they're informal, editorially-chosen bands meant to give you a sense of scale, and our own terms of use say exactly that. The Visual and Verbal ladders are different: they come from a model fitted to a published score histogram, and the site names that source under every result. What they're loosely inspired by is real psychology, though, and it's worth knowing the actual research so you can calibrate your expectations honestly.

The most famous number in working-memory research is George Miller's 1956 paper “The Magical Number Seven, Plus or Minus Two,” which surveyed a range of short-term memory tasks and found human capacity clustering around seven items, give or take two. That number has been refined a lot since — later work by Nelson Cowan and others argues the real capacity for held, actively-manipulated information is closer to four chunks, with the higher “seven” figure partly reflecting rehearsal tricks and chunking strategies people apply without realizing it. Either way, the Number Memory scores above line up remarkably well with this literature: 6–7 digits is squarely “magic number” territory, and pushing past 9 or 10 generally means you've started chunking digits into pairs or groups rather than holding each one individually.

The Chimp Test, Sequence Memory and Visual Memory load working memory differently — they add a spatial or sequential-ordering component on top of the raw item count — so they aren't directly comparable to a digit-span number. Verbal Memory swaps the material for words, and N-Back swaps the whole question, asking how well you replace what you hold rather than how much of it you can keep. That's why all six tests exist side by side: they're a rough cross-section of different facets of the same broader cognitive system, not six ways of measuring one thing.

Why none of this is a clinical measurement

It's worth being direct about what these games are and aren't. They're not validated psychometric instruments, they weren't administered under lab conditions, and your score is affected by screen size, sleep, caffeine, how many times you've played today, and plain luck in which cells the tiles land on. A single round tells you very little in isolation; a personal trend over dozens of rounds tells you a lot more. If you want a number to actually track, watch your own best score and your last-10 history over a few sessions rather than fixating on one run.

If your motivation is genuinely improving rather than just seeing a number go up once, the technique side of this matters more than raw repetition — our guide to improving working memory covers what's realistic to train and what isn't. And if you're curious why this test is named after a chimpanzee at all, the real research behind the name is more interesting than the meme format usually lets on.

Curious how the scoring and difficulty scaling actually works under the hood? See how this test works for the mechanics behind all six tests. If you would rather measure something faster than memory, a reaction time test puts a comparable number on raw response speed.