What the Human Benchmark tests measure, and what they cannot tell you about IQ

0 The short answer

Human Benchmark is a free set of reaction time, memory, aim and typing games that publishes distributions of visitor results, and none of its scores can be converted to an IQ because it has no norms for a defined population. Its games do sample real abilities that normed batteries also measure (reaction time falls under speed, and sequence, number and visual memory under working memory), and the research linking those task families to intelligence is genuine but modest for any single task. The normed counterparts are Digit Span, Alphanumeric Sequencing, Visual Sequence, Coding and Symbol Search, which ACIS opens in a free trial with no card, while its paid forms are Quick (15 dollars, about 45 minutes), Optimized (30 dollars, about 110 minutes) and Full Scale (50 dollars, about 175 minutes), with prices read on October 28, 2026 and subject to change. ACIS is one option, not the only one, and its decisive limit is stated plainly: it is online and unsupervised, it is not a clinical instrument, and it is offered in English only.

1 What Is Human Benchmark, and What Does It Say About Itself?

Human Benchmark is a free website of short browser games, and its own pages describe each one as a measurement of a single skill, never as an intelligence test. On October 28, 2026 the eight tests covered here were the reaction time test, the sequence memory test, the aim trainer, the number memory test, the verbal memory test, the chimp test, the visual memory test and the typing test. None of the eight pages describes its result as an IQ, and none states an age norm. The site shows a statistics section for each game, and this page uses only the written statements on those pages, not values read off the charts.

Each page tells you what the task is. The reaction time page asks you to click when a red box turns green and says that it is a simple tool to measure your reaction time. The sequence memory page says that buttons light up, that you press them in order, that the pattern gets longer every time you finish it, and that a mistake ends the test. The number memory page states that the average person can only remember 7 digit numbers reliably, and that mnemonic techniques can do much better. The verbal memory page describes words shown one at a time, to which you answer SEEN or NEW, with three strikes ending the run and the score being how many turns you lasted. The visual memory page describes tiles that flash white, which you must select again after they reset, with three lives. The aim trainer asks for 30 target clicks and describes itself as testing reflexes and hand-eye coordination. The typing test reports words per minute multiplied by accuracy.

The chimp test deserves its own reading, because its page makes a claim about research. It says the test is a variant of a task made famous by a study in which chimpanzees consistently outperformed humans, that some chimpanzees remembered 9 digits more than 90 percent of the time, and that the human version starts at 4 digits, adds one each turn and ends after three strikes. A later section of this page checks that claim against the published record.

What the site is, then, is a set of well designed single-skill games with global distributions of visitor results. What it is not is a norm referenced test, which means a test whose scores are interpreted against a defined reference group with published sampling and age bands. That distinction is not a criticism of the site. A thermometer and a bathroom scale are both useful instruments that nobody expects to report blood pressure. The question this page answers is which abilities each game touches, how much each of those abilities has to do with IQ, and where a normed instrument is the better tool. The page on what IQ actually measures gives the wider frame.

2 How Do the Real Options Compare for Measuring These Abilities?

A game score, a practice test and a normed battery answer different questions, and the right choice depends on whether you want to play, prepare for a screening or compare yourself with a defined population. The table lists the realistic options for someone who arrives from Human Benchmark. Prices for ACIS were read on its home page on October 28, 2026 and can change. The cost of a psychologist administered battery is set by each practice, so the table points to the page that breaks it down instead of quoting one figure.

Two rows deserve a note. A free general screener is the closest competitor to the games in price, and the page that ranks the best online IQ tests explains how to tell a documented one from an undocumented one. A practice test built for job screening can be worth taking if you have a screening coming, but it is practice for a format and not an IQ test. In the Custom Test the Full Scale IQ needs the Full Scale battery, and at 10 subtests the Full Scale form costs the same as picking them one by one. The ACIS technical manual holds the documentation behind the ACIS rows.

3 Which Ability Does Each Human Benchmark Test Sample in the CHC Model?

Each Human Benchmark game samples one narrow ability, and the Cattell-Horn-Carroll (CHC) model gives those abilities names that do not depend on the game's marketing. The CHC model organizes cognitive abilities in layers, with a general factor at the top, broad abilities such as fluid reasoning (Gf), crystallized knowledge (Gc), working memory (Gwm), visual processing (Gv) and processing speed (Gs) below it, and narrow abilities such as memory span underneath them. The mapping below is our classification of what each task demands. Human Benchmark does not classify its games this way, and the nearest normed counterpart is a similarity of demand, not an equivalence of scores. The broader page on cognitive domains in IQ testing explains how a battery groups subtests under these abilities.

Three points follow from the table. First, four of the eight games sit in the memory span family, so they overlap heavily and will tend to agree with one another more than with the others. Second, three of them (reaction time, aim and typing) depend on how fast the hand and the device respond, which is a different thing from how fast a person processes symbols in a clerical task. Third, no game samples reasoning or knowledge, the abilities that carry most of the weight in a Full Scale IQ. The page on what IQ measures describes that weighting, so a person can be quick on every game here and still have a flat profile on the abilities the games do not touch.

The ACIS subtest pages describe the normed counterparts in the same terms. The Visual Sequence page describes temporary storage for ordered visual or spatial information, which the page says should be read beside digit span and alphanumeric sequencing to see whether a difference is about format or about the broader ability. The Symbol Search page describes rapid visual comparison and decision speed, and the Coding page describes rapid symbol association and sustained output, with a caution that motor fluency and comfort with timed work can influence the score.

4 What Does Research Say About Reaction Time and IQ?

Reaction time correlates with intelligence test scores in people, but the correlation is moderate, it depends on how complex the task is, and the relationship was measured with equipment that a browser game does not replicate. The best single anchor is a population study. Deary, Der and Ford (Intelligence, 2001) tested about 900 people aged 56 who were broadly representative of the Scottish population, using a purpose built device in the style of Hick's choice reaction experiments. Scores on the Alice Heim 4 test of general mental ability correlated negative .31 with simple reaction time, negative .49 with four-choice reaction time and negative .26 with the variability of reaction times. The negative sign means that people with higher test scores responded faster. Squaring those correlations gives about 10 percent and 24 percent of shared variance, which is our arithmetic on the published figures and not a statement by the authors.

The pattern across the wider literature is the same in kind. Sheppard and Vernon (Personality and Individual Differences, 2008) reviewed 172 studies with 53,542 participants in total, and they report that measures of intelligence are significantly correlated with mental speed, and that for some measures the relationship tends to strengthen as the speeded task becomes more complex. A single click on a color change sits at the least complex end of that range, which is why the research tradition leans on choice reaction tasks and why the Human Benchmark click is the weakest of the speed measures to read as a sign of general ability.

The same Human Benchmark page that offers the test is candid about what its milliseconds contain. It states that the result is affected by the latency of the computer and monitor, that a fast computer and a low latency monitor will improve the score, that an average human reaction time may fall between 200 and 250 milliseconds while the computer could add 10 to 50 milliseconds, and that some modern televisions add as much as 150 milliseconds. Its statistics page reports more than 81 million recorded clicks, a median of 273 milliseconds and a mean of 284 milliseconds, says that recorded times have grown slightly slower over the years because of input and display technology, notes that mobile taps tend to be slower than clicks, and gives 30 milliseconds as a typical lag for a desktop or laptop. A figure for typical lag is useful and honest, and it also shows the measurement problem: two people with identical nervous systems and different screens will get different scores.

Timing research supports the caution. Reimers and Stewart (Behavior Research Methods, 2015) measured presentation and response timing across 19 computer systems with specialist hardware and found that variability within a system was generally small, that variability of measured response times across systems was somewhat larger, and that online experiments could detect differences between conditions in most within-subject and between-subject designs. They also noted concerns about using some correlational or longitudinal designs online. A person comparing a score with a stranger's score is making a between-systems comparison, which is the harder case.

So the honest reading is narrow. Reaction time is a legitimate research variable that relates to intelligence at the group level, a Human Benchmark click is a noisy version of it, and a millisecond figure says little about general ability for one person. The page on the processing speed test already sets out why a reaction time result is not a Processing Speed Index, so this page does not repeat that argument. The page on processing speed and aging covers the part of the picture that changes most across adulthood. A related speeded attention task, the Stroop color-word test, has its own evidence on the Stroop test and IQ page.

5 Do Sequence Memory and Visual Memory Predict IQ?

Sequence memory and visual memory measure how much ordered or spatial information a person can hold for a few seconds, and that capacity is related to reasoning ability without being the same thing. Both games grow the load until the player fails, so the score is the longest level reached. That is the structure of a span task. The sequence memory game resembles a family of block-tapping tasks used in neuropsychology for decades, and the Corsi Block-Tapping Task has a standardized administration and scoring procedure with published normative data: Kessels and colleagues (Applied Neuropsychology, 2000) provide a percentile distribution and cutoffs from 70 healthy participants and 70 patients with cerebral lesions. The comparison matters because it shows what a normed version of this kind of task looks like. The Human Benchmark version has the format but publishes no reference group.

How large is the capacity being measured? Cowan (Behavioral and Brain Sciences, 2001) reviewed a wide range of data and concluded that a single central capacity limit averaging about four chunks is implicated when rehearsal and grouping are blocked, which is lower than the classic figure of about seven. Luck and Vogel (Nature, 1997) showed that people could retain information about only four colors or orientations in visual working memory at one time, and that they could retain both the color and the orientation of four objects, which indicates that visual working memory stores integrated objects rather than individual features. Those results explain why games of this kind get hard quickly after four or five items, and why Human Benchmark's own number memory page speaks of 7 digits as typical while mnemonic techniques can do better. The two framings do not contradict each other: a span of seven digits usually relies on chunking, and a game with unfamiliar spatial positions blocks it.

What does this capacity have to do with IQ? Kane and colleagues (Journal of Experimental Psychology: General, 2004) tested 236 participants on three span tests each of verbal working memory, visuospatial working memory, verbal short-term memory and visuospatial short-term memory, along with reasoning tests and fluid intelligence. They report that the working memory tasks largely reflected a domain-general factor, that this working memory construct was a strong predictor of fluid intelligence and a weaker predictor of domain-specific reasoning, and that the short-term memory tasks, built from the same stimuli, were much more domain specific and showed the reverse pattern. A span game that only asks for storage is closer to their short-term memory tasks than to their working memory tasks, which require holding information while doing something else. That is our reading of the design, and it explains why the page on working memory testing distinguishes plain span from manipulation.

For a reader who wants a normed visual span, the Visual Sequence subtest is the counterpart, and the sibling page on the visual memory test covers the wider family of visual memory measures. A different skill, spotting the rule that generates a sequence, is covered on the page about the pattern recognition test, because remembering a pattern and inferring its rule are different abilities with different relationships to fluid reasoning.

6 Is the Chimp Test an IQ Test, and Do Chimpanzees Really Beat Humans?

The chimp test is a memory span game with a memorable origin story, and the published record behind that story is narrower than the headline: one juvenile chimpanzee, three students who received no practice, and a reply showing that practiced students outperformed the chimpanzee. The Human Benchmark page, read on October 28, 2026, calls the test a test of working memory made famous by a study in which chimpanzees consistently outperformed humans. The study is Inoue and Matsuzawa (Current Biology, 2007), a two page report titled "Working memory of numerals in chimpanzees." This page relies on the published reply for the design details, because the reply describes them directly.

The reply is Cook and Wilson (Psychonomic Bulletin and Review, 2010). It states that in the earlier report a juvenile chimpanzee outperformed three university students on memory for briefly displayed digits in a spatial array, and that the original authors described the abilities as extraordinary and likened the chimpanzee's performance to eidetic memory. Cook and Wilson point out that the chimpanzee had received extensive practice on a version of the task that was not under time pressure, while the human participants had received none. Their own report is that after adequate practice, two university students substantially outperformed the chimpanzee, and they conclude that there is no evidence for a superior or qualitatively different spatial memory system in chimpanzees. The page on photographic memory covers the wider evidence on eidetic claims in people.

Two things follow for anyone playing the game. First, the task in the study was a masked, time pressured display of numerals in a spatial array, and what the Human Benchmark page describes is a variant that starts at 4 digits, adds one per turn and ends after three strikes. The comparison between the two is a resemblance of concept, and the page itself uses the word variant. Second, the claim that some chimpanzees remembered 9 digits more than 90 percent of the time is made on the Human Benchmark page, and this page did not verify it against the full text of the original report. Treat it as the site's summary and not as an established figure. Neither point is a problem for the game as entertainment. Both matter if the question is whether doing well on it says something about intelligence.

On that question, the relevant literature is the one on working memory and intelligence. Ackerman, Beier and Boyle (Psychological Bulletin, 2005) meta-analyzed 86 samples relating working memory to intelligence and report that the average correlation between true-score estimates of working memory and general intelligence is .479, which they describe as substantially less than unity. They also distinguish short-term memory from working memory in a supplemental meta-analysis. A span game sits at the short-term memory end of that distinction, as the earlier section on Kane and colleagues explained. The reasonable summary is that memory span carries real information about cognitive ability and leaves most of the variance in intelligence unexplained.

Readers who search for the "chimp test average" want a single number, and the honest answer is that any distribution the page shows is a distribution of visitor results, not a norm. The people who play are people who chose to play, on devices of their choice, with as many attempts as they wanted. The score at the middle of that crowd tells you where you rank among that crowd. It does not tell you where you would rank among adults of your age in a sample designed to represent them.

The normed counterparts are different in format and similar in what they sample. The Digit Span subtest presents digits as audio, so it needs working sound, and its page describes forward and backward recall, which separates plain storage from manipulation. The Alphanumeric Sequencing subtest also presents its items as audio and asks the person to hold mixed letters and numbers and reorder them by a rule, which its page describes as updating, organization and controlled mental rearrangement. Both are part of the working memory index, with Visual Sequence as the third subtest.

7 What Does the Verbal Memory Test Measure?

The verbal memory test measures recognition of words you have already seen, which is a different demand from holding a short list in mind, even though the page describes it as a short-term memory task. The page says that the test measures how many words you can keep in short-term memory at once, and that the score is how many turns you lasted. Because the same words come back after many other words have passed, the player is judging familiarity against a growing set of items seen earlier, and interference between similar words rises with the length of the run. In the CHC model, recognition and recall of recently learned material belong to long-term storage and retrieval (Glr), which is separate from working memory. That reading is ours and not the site's.

The consequence for an IQ comparison is that this game has no close counterpart among the ACIS subtests. The nearest verbal tasks in the battery, such as Vocabulary and Similarities, measure knowledge and verbal reasoning, which are crystallized abilities (Gc), not memory. A reader who wants to understand how memory in general relates to scores can use the page on IQ and memory, which separates the memory abilities that load on the working memory index from those that do not.

This page does not quote a correlation between performance on this game and IQ, because none of the sources checked for this page studied this specific format, and a correlation for a different word memory task would not transfer. The language of the words also matters. The game is in English, and performance on word tasks depends on how familiar the words are, so a player who learned English as an additional language is measured partly on vocabulary exposure. The ACIS technical manual makes a parallel point about its own verbal and knowledge tasks, which are affected by English proficiency and educational exposure.

8 What Do the Aim Trainer and the Typing Test Measure?

The aim trainer and the typing test measure how quickly and accurately a practiced hand can execute a movement, and the result depends heavily on the device in the hand and the practice behind it. The aim trainer page says that the test tests reflexes and hand-eye coordination, that the score and average time per target appear after 30 targets, and that the test is best taken with a mouse or tablet screen because trackpads are difficult to score well with. That last sentence is an admission that the input device changes the score. In CHC terms the closest ability is psychomotor speed, the speed of movement, and the game also has a visual search component.

The typing test uses the standard words per minute formula, which divides characters by five and then by time, and records words per minute multiplied by accuracy. Typing looks simple, and the research shows it is not. Salthouse (Psychological Bulletin, 1986) reviewed typing research using a four-component model in which text is grouped into familiar chunks, decomposed into characters, converted into movement specifications and then executed. He used it to integrate 29 empirical phenomena, including differences associated with increasing skill, and he describes typing as an intricate and complex interaction of perceptual, cognitive and motoric processes. That is a description of a skill, and skill comes from practice.

A typing score therefore reflects keyboard hours as much as any underlying ability, and the same applies to a person who has spent years aiming a mouse. The normed subtest that comes closest in demand is Coding, because it samples rapid symbol association and sustained output, but its own page warns that motor speed and comfort with timed work can influence the score and that it is better read beside Symbol Search than alone. Neither Human Benchmark game has a reference group, so a fast result tells you that you are quick on your device and says nothing about your standing among adults your age.

9 Why Can a Human Benchmark Score Not Be Converted to an IQ?

A Human Benchmark score cannot be converted to an IQ because an IQ is a position in a defined reference population, and the site has no such population, no control over retries, no control over devices and no age bands. An IQ on the usual scale has a mean of 100 and a standard deviation of 15, which is a way of saying how far a person sits from the average of a reference group. The page on how IQ scores are normed walks through that logic. Without the group, the scale has nothing to attach to, and any formula that mapped game results onto it would be invented. This page does not offer one, and no honest source can.

There are four separate reasons, and each would be enough on its own.

The reference group is undefined

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), which the last content section returns to, state in Standard 5.8 that norms, if used, should refer to clearly described populations, and that those populations should include the people with whom test users will ordinarily want to compare examinees. Human Benchmark publishes distributions of what visitors recorded, and a visitor population is defined by who chose to visit. It has no stated age bands, and the reaction time statistics page says nothing about who the players are beyond the number of clicks. The median of 273 milliseconds is the middle of 81 million clicks from whoever clicked, on whatever device they used.

Retries are open, and retest effects are large

The site lets you play again and offers a saved history, so a visitor's best or latest score reflects practice. The retest literature quantifies how much practice matters. Scharfen, Peters and Holling (Intelligence, 2018) meta-analyzed 174 samples from 122 studies, with 153,185 participants in all, and found significant retest effects and no further score gains after the third test administration. For working memory capacity tests specifically, Scharfen, Jansen and Holling (Psychonomic Bulletin and Review, 2018) report a weighted average effect of g = 0.28 from the first to the second administration, with a significant increase in effect sizes up to the fourth administration and differences between paradigms. In plain terms, a person who plays ten times is not the person who played once. The Standards say in the comment to Standard 6.1 that retest policies should be established and communicated as part of the conditions for standardized administration, and a game that invites unlimited play has no such policy because it is not trying to produce a norm referenced score.

Devices change the milliseconds

The reaction time page says its result is affected by the latency of the computer and monitor, and the aim page says trackpads are difficult to score well with. The research on online timing found cross-system variability in response times that was somewhat larger than within-system variability (Reimers and Stewart, 2015, discussed above). A normed subtest controls what it can: it standardizes the instructions, the scoring rules and the comparison group, and where it cannot control the device, it says so. That is the difference between a measurement with a known source of error and a score with no stated one.

One short task is a thin measurement

A single game result is one sample of one narrow ability, taken in a minute. Reliability, which is the consistency of a score, generally rises with the number and breadth of items and tasks that feed it, and a battery combines several subtests into an index for that reason. The page on reliability and validity in IQ testing explains the two terms and how to read them. The practical point is that a single game cannot give you a confidence interval, because no norm and no reliability estimate stand behind it.

10 What Can a Human Benchmark Profile Tell You, and What Should You Not Infer?

A Human Benchmark profile can show you which kinds of task feel easy or hard for you and how you improve with practice, and it should not be read as a ranking of your intelligence. That is a genuine use. Playing the memory games repeatedly shows how quickly practice raises a score, which links to the retest evidence above. Comparing your own sequence memory with your own reaction time shows an uneven pattern that is yours. The limit is the step from a pattern in a game to a statement about ability, which needs a reference group and a measure of error.

Group correlations are not individual diagnoses either. The correlations reviewed above (negative .31 and negative .49 for reaction time, and .479 for working memory with general intelligence) describe how two measurements move together across many people. They do not say that a fast person is bright or a slow person is not, and the same studies show wide scatter around the trend. Many people have a very fast reaction time and an average reasoning score, or the reverse, and an uneven profile is common. The page on index score differences describes how often normed index scores differ from one another in a standardization sample, which is why one low or high result does not define a person.

The same caution applies to the questions people post in forums about whether a game result proves a particular IQ. This page did not sample or quote any forum thread. The sources here support a narrower answer: a result shows how you performed on that task, on that device, on that day, compared with whoever else played.

Age matters as well. Adult norms on ACIS cover ages 16 to 90, and the page on the IQ test for teenagers addresses what changes at 16 for younger readers. Absolute performance on speed and span tasks declines across adulthood while standing among age peers stays comparatively stable, which is stated on the ACIS Digit Span and Symbol Search pages, so a raw span or millisecond score is only interpretable against an age reference.

The pattern repeats for other popular games. A skill that is practiced for years, such as chess, produces performance that reflects both ability and training, and the sibling page on the chess IQ test takes up the same question for that game.

11 How Do Normed Scores Work, and How Should You Take Them Online?

A normed score places your raw performance on a scale defined by a reference group, reports it with an interval of uncertainty, and depends on conditions you can control at home. The scale is the standard one: indices and the Full Scale IQ have a mean of 100 and a standard deviation of 15, scaled subtest scores run from 1 to 19 with a mean of 10 and a standard deviation of 3, and the report gives percentiles and a 95 percent confidence interval. The confidence interval is the range within which the true score probably lies, and it exists because every score contains error. An index built from several subtests is a more precise estimate than any one subtest, and the page on the IQ score chart shows how the levels map to percentiles. A scaled score of 10 is average, and the same raw span can be average at one age and above average at another.

Conditions matter more online than in a clinic, and the Standards are direct about this. The comment to Standard 6.1 says that test developers should specify the standardized administration conditions that support intended uses, and that technology and the Internet have made it possible to administer tests in settings where administration conditions may not be strictly controlled. For ACIS the practical list is short:

• Use a device and input method you can keep constant for the whole session, and avoid a trackpad for speeded subtests if you can use a mouse.
• Turn on working sound before Digit Span, Alphanumeric Sequencing or Arithmetic, because those subtests present their items as audio. A session without sound cannot complete them.
• Take speeded subtests such as Coding and Symbol Search in a quiet room with no interruptions, because the timer does not wait.
• Use the breaks. ACIS allows breaks, saves progress between subtests and gives 30 days to complete a form, so a long battery does not need to be taken in one sitting.

The free trial, which needs no card, opens five subtests when started from the Full Scale form: Digit Span, Alphanumeric Sequencing, Visual Sequence, Coding and Symbol Search, read on the published site on October 28, 2026. They are the normed counterparts of several Human Benchmark tasks, since Digit Span and Alphanumeric Sequencing resemble number memory and the chimp test in what they sample, Visual Sequence resembles sequence and visual memory, and Coding and Symbol Search sample speeded performance of the kind a reaction game only approaches. Finished trial subtests are kept if you later unlock a form that includes them. The trial is offered once per person, and the site refuses a second trial from the same email address or the same browser.

For a reader who only wants those five abilities and no more, the Custom Test costs 5 dollars per subtest, so the same five subtests would cost 25 dollars. That is our arithmetic on the published price, and the Custom Test returns a complete index for any domain you finish, but the Full Scale IQ needs the Full Scale battery. The page on the free IQ test explains what the free trial includes and what it does not, and the page on IQ testing for adults is the wider entry point for someone who wants a full score.

12 Where ACIS sits, and what a buyer should do

ACIS sells a competing paid assessment, so this section is a disclosure: it is a normed online battery with real limits, and the right choice depends on what you need the score for. The ACIS strengths are the ones the comparison table lists: adult norms for ages 16 to 90, a Full Scale IQ and six indices with a 95 percent confidence interval, one time prices from 15 to 50 dollars without a subscription, breaks and 30 days to finish, and a free trial with no card. The technical manual is the place to check the documentation, and this page deliberately quotes nothing from it beyond what the home page publishes.

The limits are equally plain. ACIS is administered online without a supervisor, so conditions are the test taker's responsibility. It is not a clinical or diagnostic instrument, and a score should not be used for hiring, for school accommodations or for admission to high IQ societies. It is offered in English only, and the manual says verbal and knowledge tasks are affected by English proficiency and educational exposure. If you need a diagnosis, a school accommodation or an evaluation accepted by an institution, a licensed psychologist with a supervised battery is the correct route, and the page on where to take an IQ test lists the options in person and online.

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), which are open access on testingstandards.net, set the frame for reading any score. Standard 1.1 says the test developer should set out clearly how scores are intended to be interpreted and used, should delimit the population the test is intended for, and should describe the construct being assessed. Standard 5.8 asks for norms that refer to clearly described populations. The comment to Standard 6.1 asks users to be aware of the implications of less controlled administration conditions. Applied to this page, a game with no population and no conditions meets none of those expectations for an IQ interpretation, and an online battery meets the norm and construct expectations while carrying the administration caveat openly.

The American Psychological Association's Ethical Principles of Psychologists and Code of Conduct points the same way for professionals. Standard 9.02(b) says psychologists use assessment instruments whose validity and reliability have been established for the population tested, and that where they have not been established they describe the strengths and limitations of results. Standard 9.06 says that when interpreting results, psychologists take into account the purpose of the assessment, test factors and characteristics of the person assessed that might reduce the accuracy of interpretation, and indicate significant limitations. A reader without a clinician can borrow the habit: write down the purpose, the conditions and the limits next to any score.

A concrete next step depends on the reader. If you enjoy the games, keep playing them and treat the results as a log of your own practice. If you want to know where your working memory and speed stand against adults, start the free trial, finish the five subtests, and read the scaled scores with their percentiles. If you want a Full Scale IQ, choose Quick, Optimized or Full Scale after reading the example report for each form, and remember that the refund policy reads: "Full refund within 5 days if the report does not deliver the features described at checkout or a technical issue prevents you from accessing your results." Prices were read on October 28, 2026 and can change.

13 Sources Behind This Page

Every figure on this page traces to one of the following sources, each linked at the point where it is used, and the one derived figure is labeled as our arithmetic in the text. Prices and Human Benchmark statements were read on October 28, 2026 and can change.

• Human Benchmark. Test pages for reaction time and its statistics, sequence memory, aim trainer, number memory, verbal memory, chimp test, visual memory and typing. humanbenchmark.com, read October 28, 2026.
• Sheppard LD and Vernon PA. Intelligence and speed of information-processing: A review of 50 years of research. Personality and Individual Differences, 2008, volume 44, issue 3, pages 535 to 551.
• Deary IJ, Der G and Ford G. Reaction times and intelligence differences: A population-based cohort study. Intelligence, 2001, volume 29, issue 5, pages 389 to 399.
• Inoue S and Matsuzawa T. Working memory of numerals in chimpanzees. Current Biology, 2007, volume 17, issue 23, pages R1004 to R1005.
• Cook P and Wilson M. Do young chimpanzees have extraordinary working memory? Psychonomic Bulletin and Review, 2010, volume 17, issue 4, pages 599 to 600.
• Cowan N. The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 2001, volume 24, issue 1, pages 87 to 114.
• Luck SJ and Vogel EK. The capacity of visual working memory for features and conjunctions. Nature, 1997, volume 390, issue 6657, pages 279 to 281.
• Kane MJ, Hambrick DZ, Tuholski SW, Wilhelm O, Payne TW and Engle RW. The generality of working memory capacity: A latent-variable approach to verbal and visuospatial memory span and reasoning. Journal of Experimental Psychology: General, 2004, volume 133, issue 2, pages 189 to 217.
• Ackerman PL, Beier ME and Boyle MO. Working memory and intelligence: The same or different constructs? Psychological Bulletin, 2005, volume 131, issue 1, pages 30 to 60.
• Kessels RPC, van Zandvoort MJE, Postma A, Kappelle LJ and de Haan EHF. The Corsi Block-Tapping Task: Standardization and normative data. Applied Neuropsychology, 2000, volume 7, issue 4, pages 252 to 258.
• Salthouse TA. Perceptual, cognitive, and motoric aspects of transcription typing. Psychological Bulletin, 1986, volume 99, issue 3, pages 303 to 319.
• Scharfen J, Peters JM and Holling H. Retest effects in cognitive ability tests: A meta-analysis. Intelligence, 2018, volume 67, pages 44 to 66.
• Scharfen J, Jansen K and Holling H. Retest effects in working memory capacity tests: A meta-analysis. Psychonomic Bulletin and Review, 2018, volume 25, issue 6, pages 2175 to 2199.
• Reimers S and Stewart N. Presentation and response timing accuracy in Adobe Flash and HTML5/JavaScript Web experiments. Behavior Research Methods, 2015, volume 47, issue 2, pages 309 to 327.
• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing. AERA, 2014, Standards 1.1, 5.8 and 6.1.
• American Psychological Association. Ethical Principles of Psychologists and Code of Conduct, including 2010 and 2016 amendments, Standards 9.02 and 9.06, read October 28, 2026.

14 Frequently Asked Questions