Data study
Average reaction time, memory span and attention scores: data from 775 players
What people actually score the first time they take ten classic cognitive tests, from 5,103 sessions scored on our server. Every figure carries the number of players behind it.
Recomputed hourly · 13 July 2026 to 10 November 2026 · 9 of 10 tests with enough data
Key findings
- Reaction time. Collecting data since the timing fix on 4 November 2026: 13 players so far, not enough players yet to publish a median (the floor is 20).
- Stroop test. The median first attempt is 60 focus score out of 100; the top 10% scored 76 or more and the bottom 10% 52 or less (387 players). Players with 3+ runs improved by a median of +15 focus score out of 100 by their best run (71 players).
- Flanker test. The median first attempt is 75 attention score out of 100; the top 10% scored 92 or more and the bottom 10% 60 or less (369 players). Players with 3+ runs improved by a median of +8 attention score out of 100 by their best run (50 players).
- Digit span. The median first attempt is 6 digits recalled; the top 10% scored 8 or more and the bottom 10% 0 or less (426 players). 11% of first attempts scored 0, which means the player missed both sequences at the shortest length. Players with 3+ runs improved by a median of +2 digits recalled by their best run (92 players).
- Backward digit span. The median first attempt is 6 digits recalled backwards; the top 10% scored 8 or more and the bottom 10% 0 or less (208 players). 14% of first attempts scored 0, which means the player missed both sequences at the shortest length.
- Corsi block span. The median first attempt is 5 blocks recalled in order; the top 10% scored 6 or more and the bottom 10% 3 or less (142 players). Players with 3+ runs improved by a median of +1 blocks recalled in order by their best run (44 players).
- N-back (2-back). The median first attempt is 13 memory score out of 100; the top 10% scored 88 or more and the bottom 10% 0 or less (210 players). 47% of first attempts scored 0, which means no more correct matches than false alarms (pressing on every letter scores 0 too). Players with 3+ runs improved by a median of +25 memory score out of 100 by their best run (27 players).
- Mental arithmetic. The median first attempt is 16 correct answers; the top 10% scored 19 or more and the bottom 10% 11 or less (355 players). Players with 3+ runs improved by a median of +2 correct answers by their best run (51 players).
- Numerosity. The median first attempt is 8 levels solved; the top 10% scored 11 or more and the bottom 10% 5 or less (260 players). Players with 3+ runs improved by a median of +2 levels solved by their best run (67 players).
- Mental rotation. The median first attempt is 6 rotations spotted; the top 10% scored 10 or more and the bottom 10% 1 or less (143 players). 10% of first attempts scored 0, which means no shape pair judged correctly. Players with 3+ runs improved by a median of +3 rotations spotted by their best run (33 players).
First attempts, test by test
Each player counts once per test, on their first attempt. "Top 10%" is the strongest tenth whichever way the scale runs, so on reaction time it is the fastest — the smallest number. The bar runs from zero: its thin line covers the middle 80% of first attempts, the thick part the middle half, and the mark is the median.
| Test | Bottom 10% | Median | Top 10% | Spread | Based on |
|---|---|---|---|---|---|
| Reaction time milliseconds · lower is better | Collecting data since the timing fix on 4 November 2026 — not enough players yet (the floor is 20) |
13 players 36 sessions |
|||
| Stroop test focus score out of 100 | 52 | 60 | 76 |
387 players 831 sessions |
|
| Flanker test attention score out of 100 | 60 | 75 | 92 |
369 players 636 sessions |
|
| Digit span digits recalled | 0 | 6 | 8 |
426 players 858 sessions |
|
| Backward digit span digits recalled backwards | 0 | 6 | 8 |
208 players 280 sessions |
|
| Corsi block span blocks recalled in order | 3 | 5 | 6 |
142 players 370 sessions |
|
| N-back (2-back) memory score out of 100 | 0 | 13 | 88 |
210 players 383 sessions |
|
| Mental arithmetic correct answers | 11 | 16 | 19 |
355 players 596 sessions |
|
| Numerosity levels solved | 5 | 8 | 11 |
260 players 789 sessions |
|
| Mental rotation rotations spotted | 1 | 6 | 10 |
143 players 324 sessions |
|
Does practice help?
For players who took the same test at least 3 times: how far their best run ended up above their first, as a median, and on which attempt the best run usually came. A test needs 10 such players before a figure appears.
| Test | Gain, own unit | Gain, percentiles | Best on attempt | Repeat players |
|---|---|---|---|---|
| Reaction time | fewer than 10 repeat players yet | 5 | ||
| Stroop test | +15 focus score out of 100 | +28 | 3 | 71 |
| Flanker test | +8 attention score out of 100 | +9 | 2 | 50 |
| Digit span | +2 digits recalled | +20 | 3 | 92 |
| Backward digit span | fewer than 10 repeat players yet | 9 | ||
| Corsi block span | +1 blocks recalled in order | +20 | 3 | 44 |
| N-back (2-back) | +25 memory score out of 100 | +23 | 3 | 27 |
| Mental arithmetic | +2 correct answers | +19 | 3 | 51 |
| Numerosity | +2 levels solved | +20 | 3 | 67 |
| Mental rotation | +3 rotations spotted | +35 | 3 | 33 |
Part of any gain is real skill and part is simply knowing the interface the second time; this data cannot separate the two. Percentile gains use the percentile each run was given when it was played.
Phone or computer?
Median first attempt by device, for tests where at least 20 players are on each side. Units are each test's own.
| Test | Phone | Computer |
|---|---|---|
| Stroop testfocus score out of 100 | 69 (70 players) | 60 (288 players) |
| Flanker testattention score out of 100 | 85 (47 players) | 72 (298 players) |
| Digit spandigits recalled | 6 (61 players) | 6 (320 players) |
| Backward digit spandigits recalled backwards | 4 (22 players) | 6 (165 players) |
| Mental arithmeticcorrect answers | 16 (38 players) | 16 (285 players) |
| Numerositylevels solved | 8 (27 players) | 8 (207 players) |
How the device is known: a game result does not record one, but every visit does (from the browser's user agent), and a visit shares its anonymous id with the games played in it. Each player is counted on the device most of their visits came from. Tablets, ties and players with no recorded visit are left out of this split — 703 of 775 players have a phone or computer on record — but stay in every other figure.
Methodology
- Server-scored. The browser reports what happened in each round; the score is computed on our server, so it cannot be edited in the page.
- First attempts only. Distributions use each player's first run of each test, before any practice. A player is a GamePrep account, or an anonymous browser id before signing up; the two are joined once someone registers.
- Floors. A test's distribution is published from 20 players and a practice figure from 10 players with 3+ runs. Below that the page says so rather than printing a number one more player would move.
- Percentiles are nearest-rank, not interpolated: every figure is a score somebody really got.
- Reaction time counts only runs since 4 November 2026, when the test started timing the press rather than the release (the older runs read about 100 ms slow).
- Excluded: personality, risk and questionnaire games, which read a style rather than a level. Crawlers and bots are filtered out of the visit records the device split relies on.
- Covered: 5,103 sessions by 775 players, 13 July 2026 to 10 November 2026.
- Who these people are: people practising for a game-based hiring assessment, not a random sample of the population and not anyone's candidate pool.
- Not a clinical or vendor norm. These are GamePrep's own players on GamePrep's own versions of the tasks. No employer or assessment vendor publishes its norms or cut scores, and nothing here is one. How a run becomes a percentile is on how we score; the same distributions for every game are on what a good score looks like.
How to cite
The aggregate figures on this page are free to reuse under CC BY 4.0 with a link back. They change hourly, so give the date you read them.
Suggested citation
GamePrep (2026). Cognitive test results from 775 players. benchkithq.com/gameprep/research/cognitive-test-statistics
Canonical URL
https://benchkithq.com/gameprep/research/cognitive-test-statistics
Frequently asked questions
What is the average reaction time online?
We do not publish one yet. The test's timing was corrected on 4 November 2026 and only results measured since then count: 13 players so far, under the floor of 20 we need before printing a median. The row on this page fills in by itself once enough people have taken the corrected test.
What is a good digit span score?
Among 426 people who took GamePrep's Digit span, the median first attempt is 6 digits recalled. A first attempt of 7 is in the top quarter and 8 in the top tenth; the bottom tenth scored 0 or less. Players who came back at least 3 times improved by a median of +2 (digits recalled) from their first run to their best (92 players).
What is a typical Stroop test score?
Among 387 people who took GamePrep's Stroop test, the median first attempt is 60 focus score out of 100. A first attempt of 67 is in the top quarter and 76 in the top tenth; the bottom tenth scored 52 or less. Players who came back at least 3 times improved by a median of +15 (focus score out of 100) from their first run to their best (71 players).
Does practice improve cognitive test scores?
On the 8 tests with enough repeat players, the median gain from a player's first run to their best ranged from +9 percentile points (Flanker test) to +35 (Mental rotation), among players with at least 3 runs. Part of that is real skill and part is knowing the interface; this data cannot separate the two.
Do people score differently on a phone than on a computer?
Stroop test: median first attempt 69 on a phone (70 players) against 60 on a computer (288); Flanker test: median first attempt 85 on a phone (47 players) against 72 on a computer (298); Digit span: median first attempt 6 on a phone (61 players) against 6 on a computer (320); Backward digit span: median first attempt 4 on a phone (22 players) against 6 on a computer (165); Mental arithmetic: median first attempt 16 on a phone (38 players) against 16 on a computer (285); Numerosity: median first attempt 8 on a phone (27 players) against 8 on a computer (207). Units are each test's own. The device is inferred from the player's visits, as described in the method.
Where would your first attempt land?
Play any of these tests free, no account, and your first run joins the figures above.