Average IQ by Country 2026: What South Korea's 106.97 Does and Does Not Show

In early October 2026 outlets such as ARY News reported that South Korea had taken first place in the world's IQ rankings, with an average of 106.97. The figure comes from the International IQ Test, a website that publishes an annual table of average scores by country, updated on 1 January 2026 from people who took its test during 2025. Unusually for a viral ranking, the site prints its sample counts next to every score and a methodology section beneath the table, so the claim can be checked against the site's own numbers. We read the page on 9 October 2026 and recomputed what could be recomputed. The ranking is real as a list of averages. What it is an average of is a narrower question.

What does the 2026 table say?

The table ranks 137 countries, each with at least 100 test-takers. At the top: South Korea 106.97, China 106.48, Japan 106.30, Iran 104.80, Australia 104.45, Russia 103.78, Singapore 103.56, Mongolia 102.61, New Zealand 102.35 and Viet Nam 102.26. The United Kingdom is 14th at 101.57, the United States 18th at 101.04, Germany 38th at 99.32 and India 50th at 98.44. At the bottom are Rwanda (86.90), Timor-Leste (86.70) and Somalia (83.84). Twenty-seven countries average above 100 and 110 below it. A further 77 countries and territories had fewer than 100 takers and are listed separately, unranked.

Two details change the headline. The page rolls territories into their sovereign country following UN practice, so Hong Kong (107.73, 20,793 takers), Macao (106.32) and Taiwan (105.80, 46,281) are inside China's 106.48. Counted on their own, Hong Kong would sit above South Korea. And the "previous year" column on the same page shows China first in the 2025 edition at 107.19, with South Korea at 106.43 and Japan at 106.40. South Korea's climb to first is a gain of 0.54 points for it and a loss of 0.71 for China, not a new leap in a nation.

Who took the test?

The 137 country rows add up to exactly 1,212,714, the sample size the page states, so the counts can be used to see who the sample is. Indonesia supplied 299,304 takers, 24.7% of everyone, with an average of 89.96. China supplied 229,918 (19.0%) and Ukraine 91,242 (7.5%, average 95.66). Those three countries are 51.2% of the sample; the top five by takers, adding Japan and Germany, are 58.6%; the top ten are 68.5%. At the other end the median country has 1,237 takers, 63 of 137 have fewer than 1,000, and 43 have fewer than 500. South Korea's 26,996 is 2.2% of the sample and, with about 51 million residents, well under one in a thousand of its people.

These are people who found an online IQ test, chose to take it and finished it. The page says as much in its own list of limitations: every participant had internet access, the test is modelled on the Raven's matrices format, a nonverbal reasoning measure, and any online test should be treated as informational rather than diagnostic. It also says participation varies with local interest in IQ tests and with how search engines surface the website in each country and language. That is a description of an audience. A random sample of a nation, the kind a national survey draws, would not depend on any of it. For how a matrix test turns raw answers into a score, see our guide to Raven's matrices scores explained.

Does the table average to the 100 it says?

The page states that the sample has a mean IQ of 100 and a standard deviation of 15. If the country rows are a complete split of that sample, the average of the country averages, weighted by each country's takers, has to equal the sample mean exactly. Recomputed from the table, it does not: the weighted average of the 137 country scores is 98.48, 1.5 points below 100. The unweighted mean of the 137 scores is 96.39. Leave Indonesia out and the weighted figure is 101.27. Last year's table, for the 125 countries that have a previous-year row (1,332,521 takers), gives 100.06 under the same calculation.

We cannot say why this year's table is 1.5 points off. The page does not state whether the 100 refers to every response before its screening for repeat attempts and bots, to a different pool of people, or to a scale that was set before the final table was cut. What it does show is how much the Indonesian crowd, at 24.7% of the sample and 89.96 on average, weighs on any global figure. The FAQ answers a related question, why most countries are below 100, by pointing to population weighting: China's large population offsets the many sub-100 countries. That argument is about population, which the page does not tabulate, so it cannot be checked here. What can be checked is that the takers, not the populations, are what produced the 100.

Do scores move when the crowd moves?

The page says averages are stable: in 84.8% of countries the score shifted by less than 2 points year over year. We counted 106 of 125, which is 84.8%, and the mean absolute change is 1.07 points. Stability within 2 points sounds reassuring until it is set against the sample counts, because the changes are not random with respect to them. Across the 125 countries, the change in a country's average has a correlation of -0.40 with the change in the logarithm of its sample size (Spearman -0.39): where the crowd grew, the average tended to fall, and where it shrank, the average tended to rise.

• 38 countries whose sample shrank by more than half gained 0.92 points on average
• 35 countries whose sample stayed within 25% either way changed by +0.14
• 18 countries whose sample grew by more than 25% lost 1.01 points
• Indonesia's sample doubled (150,041 to 299,304) and its average fell 3.22 points (93.18 to 89.96)
• Viet Nam's sample fell 69% (47,671 to 14,915) and its average rose 2.14; Romania's fell 75% and rose 2.57; Argentina's fell 68% and rose 2.72

It is not a law: Japan's sample fell 62% and its average moved only 0.10. And a correlation does not say what the cause is. It is consistent with a changing mix of people, for example a country's test traffic swinging between search visitors and social-media visitors from year to year. It sits awkwardly with the idea of a fixed national quantity measured by a steady instrument. The same caution applies to norms that age on their own; our report on why IQ norms expire covers what happens when even a carefully sampled reference group is a decade old.

How precise is a rank?

Sampling error is small for big samples and large for small ones. The standard error of an average is the standard deviation divided by the square root of the sample size, so with a standard deviation of 15 it is 0.09 points for South Korea, 0.03 for China and 0.13 for the Netherlands, but 1.3 for Malta (129 takers) and 1.26 for Somalia (141). Assuming a standard deviation of 15, a 95% interval for the smallest ranked countries is therefore about plus or minus 2.5 points, about 5 points from end to end. That is wider than the 3.3 points between 20th place (Switzerland, 100.84) and 60th (Algeria, 97.53).

The top of the table is the opposite case. South Korea and China differ by 0.49 points with a standard error on the difference of about 0.10, so the gap is not noise in the sample. That is exactly the trap: a large sample makes a gap look decisive, but it measures the people who took the test more precisely, not the country better. Ranks 21 to 31 (Georgia to Thailand) sit within 0.41 points of each other, 100.35 to 99.94, and tell you nothing about order. Our IQ bell curve shows how little of the distribution a half-point shift covers, and the IQ percentile calculator turns it into a rank: a person scoring 106.97 on a mean-100, SD-15 scale is at about the 68th percentile and one scoring 106.48 at about the 67th, so the half-point gap that decides first place is worth one percentile place.

What would a fair country comparison need?

A comparison of nations needs a sampling frame that covers the whole population, one instrument given under matched conditions, norms set on the people being compared, and a fixed date. National assessments come closer, and even they measure schooling as much as reasoning; see our reading of PISA 2025 and what it does not show about IQ. The older tables that circulate as national IQs have a different problem, a patchwork of studies and estimated values, which we set out in why country IQ rankings are weaker than the charts suggest. The International IQ Test table avoids that patchwork and substitutes another weakness, a very large but self-selected crowd. Neither is a measurement of the intelligence of a country. For the same question at state level, see what average IQ by US state rests on.

• Ask how many people stand behind a number, and how they came to take the test
• Ask whether the top countries are also the biggest contributors to the sample
• Treat any gap smaller than about 2 points between small-sample countries as noise
• Check whether the year-over-year change tracks the change in sample size
• Prefer sources that publish a sampling frame, a date and a norm group

Our IQ Metrics test reports a score and percentile against its own norm group. If you are weighing any online test, our checklist on what to check before trusting an online IQ score applies to the country tables as much as to individual results, and the page on average IQ by country explains the older figures in more detail. How far any single test can be trusted is covered in IQ test accuracy.

Common questions