The Organisation for Economic Co-operation and Development (OECD) reported on September 8, 2026, that more than 760,000 15-year-old students in 91 countries and economies took PISA 2025. Average OECD performance fell to the lowest levels recorded in the program in science, reading and mathematics, while Taiwan remained among the highest-scoring systems.

The OECD's PISA 2025 results cover science, reading, mathematics and a new computational problem-solving assessment. The cycle also collected information about students' learning environments, attitudes and use of digital tools, but those background measures do not turn the assessment into a test of any individual child.

What PISA measures in 2025

PISA assesses how students around age 15 apply knowledge to real-life problems. It is not a national school-leaving exam and is not designed to produce a report card for a particular student, school or district. Students took a two-hour assessment, with different students receiving different combinations of subjects, followed by a background questionnaire.

Science was the main domain in 2025. Reading and mathematics were minor domains, so the three sets of results provide different amounts of detail about each subject. The new Learning in the Digital World assessment examined computational problem solving and students' ability to direct their learning in a digital environment.

The scale is designed for comparisons across participating education systems, not for converting a score into a school grade. The OECD says more than 760,000 students represented about 33 million 15-year-olds in the participating systems. That broad sample makes the results useful for policy comparisons, but it does not remove the uncertainty created by sampling and response rates.

OECD averages fall across the core subjects

Between 2015 and 2025, the average reading score across OECD countries fell by 28 points and mathematics fell by 22 points. The OECD said science also reached its lowest average level in the program. Reading recorded the sharpest recent deterioration, including weaker performance in evaluating information, connecting material from different sources and thinking critically about what is read.

About one in five 15-year-old students across OECD countries performed below the baseline level in all three core subjects in 2025, up from 16% in 2022. The results describe a system-level pattern; they do not identify one cause. The report discusses digitalization, screen time, reading habits and differences in learning conditions, but a cross-country association is not proof that one of those factors caused the decline.

Taiwan stays near the top, with gaps inside the system

The OECD's Taiwan country note reports a science score of 540, a reading score of 508 and a mathematics score of 546. Those results placed Taiwan fourth in science, third in reading and fourth in mathematics. Taiwan scored 551 in computational problem solving, ranking fifth in that new domain.

Taiwan's science score rose by two points from 2022, while reading fell by seven points and mathematics by one point. Over the same period, the OECD comparison group of 23 countries that took part in both cycles fell by five points in science, 16 points in reading and 11 points in mathematics. A stable position in a ranking can therefore coexist with movement in a system's own scores.

The gap within Taiwan also matters. Students in the top quarter by socioeconomic status outperformed those in the bottom quarter by 91 science points, compared with an OECD average gap of 85 points. An international rank cannot summarize how students in different schools or households experience the education system.

How this plays out outside Taiwan

PISA is useful to readers outside Taiwan only when the international table is connected to the assessment arrangements in their own education system. The OECD publishes a country note and a data profile for each participant. Those pages are the right starting point for checking a national score, the number of schools and students represented, the subjects reported and any warning attached to the estimate. A headline rank should not be copied into a school, district or university-admissions claim because PISA samples students rather than testing every student in a country.

In the United States, the 2025 country note reports 4,625 students from 144 schools, representing an estimated 3.58 million 15-year-olds. The United States scored 502 in science, 463 in mathematics, 490 in reading and 513 in computational problem solving. These figures are directly relevant to US readers, but the OECD says the US student response rate fell below its minimum target, school participation also fell short and student exclusions exceeded the tolerance in the technical standards. The OECD therefore did not include questionnaire-based results for the United States and says possible bias in performance results cannot be ruled out. The US National Center for Education Statistics explains the US sampling and testing design, including the fact that the US did not administer the optional teacher and parent questionnaires in 2025. A US reader should treat the scores as national estimates with a stated quality warning, not as a ranking of states or school districts.

Singapore provides a contrasting case. The OECD country note reports 6,551 students in 157 schools, representing about 42,500 15-year-olds, and says the Singapore data met PISA's quality standards. Singapore scored above the OECD average in all three core subjects: 90% of students reached at least Level 2 in science, 89% did so in mathematics and 86% did so in reading. The figures show how a high national average can still coexist with an internal socioeconomic gap: the science difference between the top and bottom quarters was 91 points. For families and educators in Singapore, the relevant comparison is therefore not only whether Singapore remains near the top, but also how the distribution of results changes between cycles.

The United Kingdom and Australia are also participants listed in the OECD's 2025 coverage. The OECD reports that the United Kingdom was among the systems in the top ten across science, mathematics and reading, while Australia appears in the report's country comparisons and tables. Readers in those countries can use the OECD's country-profile links to check the national estimate and its confidence interval, then consult their own education department for curriculum or school-level policy. PISA itself does not set a national curriculum, prescribe a phone policy or provide a price for tutoring. Those questions belong to each country's education authorities and schools.

This distinction matters for digital learning. A student using an AI chatbot for homework is not automatically demonstrating computational problem solving in the PISA sense. The 2025 assessment measured how students used computational tools to work through problems, and the OECD plans further analysis of self-directed learning in a later volume. Families comparing education systems should ask what was measured, who was sampled, whether the country's technical standards were met and whether the claim concerns a test score, a classroom rule or a commercial product. Those are separate facts and should be checked on separate official sites.

Family support and digital-device use are associations

The source data report lower family-support indicators in Taiwan in 2025 than in 2022. The share of students who said a parent or guardian asked at least weekly what they had done at school fell from 62% to 59%, while the share who discussed a school problem with a parent or guardian at least weekly fell from 58% to 53%.

Across participating systems, students who reported being asked weekly about school scored about 35 points higher in science after adjustment for socioeconomic differences. That is a statistical association, not an intervention result. It cannot show that asking a question caused a particular student's score to rise.

Taiwan students also reported an average of 1.8 hours a day using digital devices for learning at school and 1.1 hours for leisure. Forty percent said they used an AI chatbot for learning at least weekly. These responses describe reported use and cannot establish that more or less device time improves learning.

Frequently asked questions

Who takes PISA?
PISA tests students roughly 15 years old who are enrolled in school. It is a sample-based international assessment, not a test taken by every student or a direct measure of an individual school.

What were Taiwan's PISA 2025 results?
Taiwan scored 540 in science, 508 in reading, 546 in mathematics and 551 in computational problem solving. Its reported ranks were fourth, third, fourth and fifth respectively.

Does a higher rank prove a better education system?
No. Results must be read with confidence intervals, sampling information, participation rates and the subject tested. A rank difference can be smaller than the uncertainty around the estimates.

Did family questions raise scores by 35 points?
No causal conclusion follows from that figure. It is an adjusted association in PISA data, not a controlled trial of family conversations.