Average IQ of Software Engineers: What the Research Actually Measured

The average IQ of a software engineer has not been measured in any study we could find. Search for it and you will be told 120, or a range of 115 to 132, or 130, usually with no source. What exists instead is smaller and more interesting: a handful of studies on how reasoning tests relate to coding tasks, an official ratings database for the job's ability demands, and a revised estimate of how well general mental ability predicts job performance at all. This article sets out what each one measured, because none of them is an average IQ.

Is there an average IQ for software engineers?

We found no representative study that gave an IQ test to working software developers, so there is no checkable average. That is "none found" rather than "none exists": we read the related-work sections of the main papers and ran searches, but did not sweep every database. The authors of a 2021 paper in IEEE Transactions on Software Engineering wrote that a study on the question was missing, and called their own study the first large-scale empirical evaluation of its kind. Their sample was undergraduates, not developers.

If you meet a claimed average, convert it to a percentile first. On a mean-100, standard-deviation-15 scale, an average of 120 would put the typical developer at about the 91st percentile and an average of 130 at about the 98th (97.7th exactly). That is a very select group, and the IQ percentile calculator returns the same figures. A claim that large needs a sample, a named test and a stated scale. None of the pages we read offered any of the three.

Where do the 120 and 130 figures come from?

The pages we read name no study. One scraped page says "according to a study by payscale" that the average software engineer's IQ is around 120; we could not find any PayScale study of IQ. The same page says computer programmers range from 115 to 132 and engineers average 126, with no link or citation. A second tier of pages leans on Wonderlic occupation tables, where "programmer 29" circulates. The Riot IQ guide traces such numbers to the 1983 Wonderlic Personnel Test Manual, but we found no copy of that manual, so the cell cannot be checked. The earliest primary evidence we found is a 1977 paper in the journal Intelligence, in which Arthur Jensen noted that Wonderlic's published norms covered 80 occupational categories. It does not list programmers.

We also found no published Wonderlic table that converts its raw score to IQ. A score of 20 is described as intended to indicate average intelligence, and a rule of thumb of twice the score plus 60 circulates. That rule would turn a 29 into 118, close to the 120 many pages print, but we cannot show that this is where the figure began. Our guides to the Wonderlic score to IQ conversion and to Wonderlic scores by job go through why the shortcut is unreliable.

What did the 2026 study of coding puzzles find?

The newest paper is Hidellaarachchi, Baltes and Grundy's "How Does Cognitive Capability and Personality Influence Problem Solving in Coding Interview Puzzles?", presented at CHASE 2026 (arXiv 2511.14367). It had 80 participants, 40 software practitioners and 40 students, who completed an online survey. Cognitive capability was measured with Baddeley's grammatical reasoning test: 64 short true-or-false statements in three minutes, scored as the percentage correct of those attempted. The authors call it a brief verbal reasoning task. It is not an IQ test and the paper reports no IQ scores.

• Reasoning test: practitioners averaged 76.1% (standard deviation 22.0) and students 70.9% (15.9). The difference was not statistically significant (p = 0.227)
• Nine interview-style questions, six on coding and three on logic: practitioners averaged 3.75 of 9 and students 5.23 (p = 0.001), so the students did better
• Across all 80 people, reasoning-test accuracy correlated with problem-solving accuracy at r = 0.41, which is about 17% of shared variance by our arithmetic. Conscientiousness correlated at r = 0.39 (95% confidence interval 0.19 to 0.56)
• The design was correlational and the authors call the findings exploratory associations. Practitioners were recruited through the authors' network and social media, then Prolific

Read it for what it is. Practitioners did not beat students on the reasoning test, and they scored lower on the interview puzzles, but nine puzzle questions are not a day's work, and the paper does not claim they measure developer skill. It supports a modest link between a short reasoning test and puzzle accuracy. It gives no average IQ for developers, and it undercuts the idea that working programmers are a uniformly high-scoring group.

How much does reasoning ability explain about writing code?

Two earlier studies are closer to real programming. Wagner and Wyrich's 2021 "Code Comprehension Confounders" gave an intelligence test to undergraduates before a code-reading task. The test was the LPS-2 battery, whose results are mapped to age-adjusted IQ values. Complete data came from 117 of 135 participants. General intelligence had a standardised coefficient of 0.312 for code comprehension, and fluid reasoning 0.299. The full model explained almost 10% of the variance and the stepwise model more than 12%. The authors call these small associations. They also note that comprehension scores were close to the ceiling (mean 17.3 of 19), which shrinks any correlation.

Prat and colleagues' 2020 paper in Scientific Reports tested 36 adults over ten 45-minute Python lessons. Their models explained 50 to 72% of the variance in learning outcomes. Fluid reasoning and working-memory capacity accounted for 34%, language aptitude for 17%, resting-state EEG for 10% and numeracy for 2%. The headline "language aptitude predicts coding" hides that reasoning and working memory explained twice as much. These were novices learning to code, not working developers. For how the two abilities differ, see our note that working memory and reasoning are not the same.

What abilities does O*NET list for software developers?

The US Department of Labor's O*NET database lists software developers (code 15-1252.00) in Job Zone Four, where most jobs need a four-year degree. On the O*NET OnLine display, which scales importance from 0 to 100, deductive reasoning and written comprehension score 69, inductive reasoning 63 and mathematical reasoning 47. The underlying database file (version 30.2) shows the ratings came from eight analysts in August 2025, so they are rough. They are importance ratings for the job, not scores of the people in it, and they fit Prat's pattern: reasoning and language count for more than arithmetic.

How well does IQ predict programming performance?

The widely quoted figure is the .51 that Schmidt and Hunter reported in 1998 for general mental ability and job performance. The paper states that .51 is the average for medium-complexity jobs, with .56 for high-level complex technical jobs and .58 for professional and managerial ones. Programmers are not classified there, and the paper contains no programmer-specific value, nor does Hunter and Hunter's 1984 review. The nearest band is a reading of ours, not the authors'.

The number has since come down. Sackett, Zhang, Berry and Lievens (2022, Journal of Applied Psychology) re-analysed prior meta-analyses and found validities .10 to .20 lower than previously thought. A later paper by Sackett and colleagues quotes their mean corrected validity of .31 for general cognitive ability and finds .22 on 21st-century data (observed correlation .16). A correlation of .31 means about 10% of the variation in job performance, and .22 about 5%. Our overview of what cognitive-ability tests at work predict covers the same revision for other jobs.

What do tech employers actually say?

Laszlo Bock, then Google's head of people operations, said in a New York Times interview reported in 2013 that brainteasers were "a complete waste of time". Another reprint of an interview credited to the Times quotes him saying the main thing Google looks for is general cognitive ability, "and it's not IQ", describing it as learning ability and the ability to process on the fly. We could not open the Times pages, so both quotes are as reported by The Billfold and HR Grapevine. The practical point is the same: employers describe ability to learn, not an IQ cutoff. Our pieces on whether IQ tests are legal for hiring and on proctored hiring tests cover the testing side.

The IQ Metrics test gives a measured result, and our guide to IQ tests for job interviews explains what employers do and do not ask. For how other jobs compare, see what IQ quant trading firms look for.

Common questions