PISA 2025 Results: Record Lows, and What They Do Not Show About IQ

The OECD published the results of its 2025 Programme for International Student Assessment, known as PISA, on 8 September 2026. Across OECD countries the average 15-year-old scored 463 in mathematics, 461 in reading and 482 in science, and the OECD said mathematics and reading reached the lowest averages it has recorded. Headlines quickly turn results like these into claims about national intelligence. PISA is not an IQ test, and the arithmetic below shows both what a 22-point fall amounts to on an IQ-style scale and why that conversion should not be quoted as a finding.

What did PISA 2025 find?

More than 760,000 students took part, representing around 33 million 15-year-olds in 91 countries and economies. The 2025 cycle focused on science, alongside reading and mathematics, and added a first-time assessment of learning in the digital world. Because that last scale is new, there is no trend to read from it yet.

• OECD averages in 2025: mathematics 463, reading 461, science 482
• Mathematics: down 22 points across OECD countries between 2015 and 2025, which the OECD describes as just over one year of learning
• Reading: down 28 points over the same period, about one and a half years of learning
• Low performers: one in five 15-year-olds across the OECD, up from 16% in 2022
• Hasty reading, as the OECD labels it, nearly doubled to 9% between 2018 and 2025

Who scored highest, and who gained?

In mathematics the highest mean was B-S-J-Z (China) at 612, ahead of Singapore at 563, Macao at 549, Chinese Taipei at 546, Japan at 525, and Korea and Hong Kong at 522. B-S-J-Z stands for Beijing, Shanghai, Jiangsu and Zhejiang, four jurisdictions, not the whole of China. In reading Singapore led at 535 with B-S-J-Z at 527, and in science B-S-J-Z led at 597 with Singapore at 560.

Among larger Western systems the table shows the United Kingdom at 488 in mathematics, 494 in reading and 511 in science, and the United States at 463, 490 and 502. Australia scored 478, 491 and 509. Estonia reached 508 in mathematics, 499 in reading and 527 in science. The OECD says Costa Rica, the Slovak Republic, Türkiye and the United Kingdom increased their science scores since 2022. The table also marks several systems with an asterisk, including Canada, New Zealand, the Netherlands, Norway and the United States, meaning one or more PISA sampling standards were not met; the table's note says caution is needed in interpreting the results for those systems.

A raw score means more once you subtract the OECD average, because the PISA scale is built so that 100 points is one standard deviation. Our arithmetic on the table, in points above the OECD average of 463 in mathematics, 461 in reading and 482 in science, is below. A gap of 100 is one standard deviation, so Singapore's mathematics lead is exactly that, while the United States sits on the OECD average in mathematics.

• Singapore: +100 in mathematics, +74 in reading, +78 in science
• B-S-J-Z (China): +149 in mathematics, +66 in reading, +115 in science
• Estonia: +45 in mathematics, +38 in reading, +45 in science
• United Kingdom: +25 in mathematics, +33 in reading, +29 in science
• Australia: +15 in mathematics, +30 in reading, +27 in science
• United States: 0 in mathematics, +29 in reading, +20 in science (sampling-standards asterisk)

What does a 22-point fall mean in IQ terms?

PISA reports scores on a scale built so that 500 is the average of OECD countries and 100 points is one standard deviation, the usual measure of how widely scores spread. Each subject's scale was set in the cycle where it was the main focus: reading in 2000, mathematics in 2003 and science in 2006. IQ scales are built the same way, with a mean of 100 and, in most modern tests, a standard deviation of 15. So you can convert: 22 points is 0.22 of a standard deviation, and 0.22 times 15 is 3.3 points. Reading's 28 points converts to 4.2.

There is an independent check on the size. The OECD equates the 22-point fall to just over a year of learning. A 2018 meta-analysis in Psychological Science by Stuart Ritchie and Elliot Tucker-Drob pooled 142 effect sizes from 42 data sets with more than 600,000 participants and found that an additional year of education raised intelligence test scores by approximately 1 to 5 IQ points. A fall worth about a year of learning and a conversion of about 3 points fall in the same neighbourhood. That is reassuring about scale. It does not turn either number into a measured change in anyone's IQ. The same caution applies to the claim that Gen Z is less intelligent than its parents; see Gen Z IQ: is Gen Z less intelligent than its parents.

The size is also worth setting beside the Flynn effect. The figure usually quoted for the twentieth-century rise in IQ test scores in industrialised countries is about three points per decade. The 22-point fall in mathematics converts to about 3.3 points over the ten years from 2015 to 2025, and reading's 28 points to 4.2, so the decline is of a similar size to the typical historical gain, in the opposite direction. That comparison has the same limits as the conversion: it lines up two different kinds of score. A country league table built from online volunteers instead of a sampling design is examined in average IQ by country 2026.

The conversion rests on assumptions. It assumes the spread of scores is close to what it was when each scale was set, which we have not checked for 2025. It treats a school test of 15-year-olds as if it measured the same thing as a reasoning test, which it does not: PISA scores respond to schooling, attendance, curriculum and student engagement. And the OECD average itself depends on which countries are in the comparison. Our guide to PISA versus IQ tests sets out the difference, and our explainer on achievement tests versus IQ tests covers why the two diverge.

Does a PISA gap tell you a country IQ?

The same arithmetic produces headlines. B-S-J-Z's mathematics mean of 612 is 149 points above the OECD average of 463, or 1.49 standard deviations. Multiply by 15 and it looks like 22 IQ points. That is how many "country IQ" tables are assembled from test scores, and it is why they mislead: the sample is 15-year-olds in a selected set of schools or jurisdictions, the tests measure taught skills, and a mean is not a norm group for the whole population. Our analysis of country IQ rankings explains why the data behind such charts is weaker than they suggest, and our page on average IQ by country shows how we handle the figures.

Falling scores also raise a different question: whether IQ norms themselves are drifting. Some countries recorded rising test scores through the twentieth century, a pattern called the Flynn effect, and several data sets show it stalling or reversing. We review that evidence in why IQ norms expire. PISA is a poor test of that question because it is not an IQ test, and because a decline in its scores has many candidate causes, schooling, attendance and screen use among them, that it was not designed to separate. Our piece on screen time and cognitive test scores in children covers what the research can and cannot say.

What should readers take from PISA 2025?

• Read the OECD average (463, 461, 482) as the benchmark, and check whether a system carries a sampling-standards asterisk
• Treat 22 points as roughly a year of learning on the OECD's own scale, not as an IQ loss
• Remember that B-S-J-Z is four jurisdictions, not a country; compare like with like
• Be sceptical of any "national IQ" figure derived from a PISA gap
• Use PISA for what it is, a snapshot of 15-year-olds' school skills, and use an IQ test for reasoning

Our IQ Metrics test reports on the standard scale, and you can see how any score converts with the IQ score converter. For another school-test release from this autumn, see our report on SAT scores for the class of 2026.

Common questions