PISA 2025 Results (Volume I)

This chapter examines student performance in the PISA 2025 assessment. PISA measures student performance as the extent to which 15-year-old students have acquired the knowledge and competencies that are essential for full participation in modern societies. The first section reports the average performance in science, reading and mathematics, as well as computational problem solving1, for each country and economy, comparing it with the performance of other countries and economies. It also explores variations in performance within and between countries and economies. The second section examines changes in performance between the 2022 and 2025 PISA assessments. It also discusses long-term trends in student performance in PISA stretching over a decade. The final section of this chapter covers two dimensions of equity in education – inclusion and fairness – and looks at changes over time among countries and economies that participated in PISA 2025 and in previous assessments.

In PISA 2025, students scored 482 points in science, on average across OECD countries (Table I.B1.2a.1). Students at this average score are at a proficiency level where they can use scientific ideas, evidence and representations to construct or evaluate explanations, simple models, investigations and conclusions when given appropriate support. They can also interpret simple data, identify strengths, limitations or flaws in models and research, and choose suitable methods or sources of evidence. They can justify their choices using relevant evidence and basic criteria for evaluating the credibility of sources (see the section: What students can do in science) (Table I.2.2).

Students in B-S-J-Z (China) (597 points) and Singapore (560 points) scored higher in science than students in all other countries/economies that participated in PISA 2025 (Table I.2.1). On average, these students are able to engage with broader and more demanding scientific ideas, more complex investigations and sources. They are expected not only to interpret or choose evidence, but also to evaluate claims, identify errors and describe flaws.

Science was the major domain in PISA 2025, but performance in mathematics and reading provides important additional context for comparing education systems across the core domains.

In mathematics, students scored 463 points on average across OECD countries. In B-S-J-Z (China), students performed the highest in mathematics (612 points), while the second-highest was achieved by students in Singapore (563 points). In reading, the ordering is reversed for students in Singapore, who had the highest mean performance (535 points), followed by B-S-J-Z (China) (527 points) – across OECD countries, students scored 461 points in reading, on average (Tables I.B1.2a.2 and I.B1.2a.3).

Table I.2.1, I.2.o1 (online) and I.2.o2 (online) show each country’s/economy’s average score and indicate which peer systems differ statistically significantly2 from each other.

In PISA 2025, all countries and economies that performed above the OECD average in science also performed above the OECD average in mathematics and reading, with four exceptions. Belgium and Lithuania performed above the OECD average in science and mathematics but not statistically different from the OECD average in reading; while Türkiye and the United States* performed above the OECD average in science and reading but not statistically different from the OECD average in mathematics. In total, 18 countries and economies performed above the OECD average in science, mathematics and reading (Table I.2.1, I.2.o1 (online) and I.2.o2 (online)).

The distance between the highest- and lowest-performing systems provides a simple snapshot of the global range of student achievement in PISA 2025. Although purely descriptive, it helps locate each system within the broader international distribution and illustrates the scale of performance differences observed across education systems.

The gap in performance between the highest- and lowest-performing countries is 124 score points in science among OECD countries and 280 points among all education systems that took part in PISA 2025 (Table I.B1.2a.1). In reading, the performance gap between the highest- and lowest-performing countries is 103 score points among OECD countries and 234 score points among all education systems that took part in PISA 2025 (Table I.B1.2a.2). In mathematics, the corresponding gap is 145 score points among OECD countries and 301 score points among all education systems that took part in PISA 2025 (Table I.B1.2a.3).

Finally, Box I.2.2 reports estimates suggesting that students around age 15 typically gain about 20 score points over one year of schooling, i.e. this number of points could be considered to be equivalent to one school year of learning. This benchmark provides useful context when assessing changes in performance over time, but it should be interpreted cautiously and not applied mechanically to cross-country differences in mean scores.

Variation in student performance, as measured by the standard deviation of scores, tends to be greater among high-performing than low-performing education systems. There is a strong correlation between average performance in science and variation in performance in science (see Figure I.2.o1 (online)). For example, Cambodia has the smallest variation in science proficiency (63 score points) while several other countries and economies whose mean performance was below the OECD average also have small variations in performance3 (Table I.B1.2a.1). However, this pattern does not hold in all cases. Among countries/economies that performed above the OECD average, B‑S‑J‑Z (China), Estonia, Lithuania and Macao (China) stand out for their comparatively small variation in performance (standard deviation under 90 score points). Similarly, among countries that performed below the OECD average, Bulgaria, Israel, Luxembourg, Malta, the United Arab Emirates and Zambia stand out for their relatively large variation in performance (standard deviation greater than 105 score points).

Another measure of variation in performance within countries is the score gap that separates the highest- and lowest-performing students within a country (i.e. inter-decile range). In science, the difference between the 90^th percentile of performance (the score above which only 10% of students scored) and the 10th percentile of performance (the score below which only 10% of students scored) is about 237 score points in all participating countries and economies, and 258 score points on average across OECD countries (Table I.B1.2a.1).

The largest differences between high- and low-achieving students in science are found in Luxembourg and the United States* (Figure I.2.1 and Figure I.2.o1 (online)). In these countries, the inter-decile range is over 290 score points, indicating that student performance in science is highly unequal among 15-year-old students. By contrast, the smallest differences between high- and low-achieving students are often found among countries and economies with low mean performance (i.e. lower than 410 points), as in Cambodia, Dushanbe (Tajikistan), Indonesia, Kosovo and the Kurdistan Region (Iraq). In these countries, the 90th percentile of the science distribution is below the OECD average score (Table I.B1.2a.1). In some lower-performing systems, smaller differences between high- and low‑achieving students may reflect a compressed performance distribution, in which relatively few students reach high levels of proficiency and even the 90th percentile remains below the OECD average.

Examining variation in performance between and within schools is important, because it shows whether performance gaps are mainly structured by school attendance or are predominantly found among students within the same school. When a large share of variation lies between schools, students’ outcomes differ substantially depending on the school they attend. When most variation lies within schools, differences in performance are found mainly among students attending the same school. This distinction helps identify whether policy efforts need to focus primarily on differences between schools, differences among students within schools, or both.

Out of the variation observed within countries in PISA 2025, 31% of the OECD-average variation in science performance is between schools (Figure I.2.2); while 69% is within schools. This means that, on average and in most countries/economies, school characteristics may play a less dominant role in explaining differences in student performance than student-level and classroom-level characteristics. These could include students’ background, dispositions and learning behaviours, as well as differences associated with grade level, student grouping within the same school.

However, the extent of between-school variation in science performance differs widely across countries and economies. In three countries and economies, between-school differences account for less than 10% of the total variation in performance (Iceland, Finland and Norway*, in ascending order). By contrast, in ten other countries (Zambia, the Netherlands*, Kenya, Bulgaria, Türkiye, Rwanda, Hungary, the Slovak Republic, Germany and G710Romania, in descending order), differences between schools account for at least 50% of the total variation in the system’s performance. This suggests that education systems in the latter group are more academically segmented, with high- and low-performing students more likely to be concentrated in different schools.

PISA defines a scientifically educated person as someone who can engage in reasoned discourse about science, sustainability and technology to inform action (OECD, 2023[1]). Such an individual is educated to be an informed, critical user of scientific knowledge throughout their life, without necessarily becoming a producer of scientific knowledge. Such individuals possess an understanding of major scientific concepts, the standard practices of scientists, the degree to which scientific claims are justified, and the social mechanisms that make scientific knowledge trustworthy. This empowers them to evaluate the credibility of scientific information, make informed personal decisions, and play a role in societal choices impacting the environment.

To successfully demonstrate these competencies, three distinguishable but related forms of knowledge are required:

• Content knowledge: An understanding of the scientifically established facts, concepts, ideas and major explanatory theories about the natural and material world.

• Procedural knowledge: A familiarity with the standard procedures and practices that scientists use to obtain reliable and valid data. This involves familiarity with the concepts underlying empirical enquiry, such as identifying variables, repeating measurements to minimise uncertainty, and knowing how to appropriately represent and communicate data.

• Epistemic knowledge: An understanding of how scientific claims are justified and the rationale behind the procedures scientists use. This requires grasping the role of observations, models, theories and arguments, as well as understanding the collaborative nature of science, the importance of peer review, and how scientific consensus establishes trustworthy knowledge.

A scientifically educated person applies their knowledge in real-world personal, local or global contexts. For such application, PISA identifies three specific competencies (see Box I.2.1):

• Explain phenomena scientifically: The ability to recognise, construct and evaluate explanations and models for natural and technological phenomena.

• Construct and evaluate designs for scientific enquiry and interpret scientific data and evidence critically: The ability to appraise how questions are investigated scientifically and to critically interpret data.

• Research, evaluate and use scientific information for decision making and action: The ability to find information, evaluate its credibility and potential flaws, and use it to inform decisions.

To assess how scientifically educated 15-year-old students are, the PISA 2025 science assessment scale is divided into descriptive proficiency levels, ranging from Level 6 (the highest) down to Level 1b (the lowest). Table I.2.2 illustrates the range of science competencies covered by the proficiency levels and describes the skills, knowledge and understanding required at each aggregated level (the full science proficiency descriptors for PISA 2025 are included in Table I.2.o7 (online), in Table I.2.o8 (online) for reading and in Table I.2.o9 (online) for mathematics).

In PISA, proficiency Level 2 is considered the baseline level of proficiency students need to participate meaningfully in society. Level 2 is regarded as the baseline level of proficiency because it marks the point at which students use scientific knowledge in ways that demonstrate scientific literacy: they pass from knowing some science to using science at a basic level.

Below this level, students may demonstrate some familiarity with scientific ideas or recognise simple information in familiar contexts. At Level 2, however, students are expected to go further: they can use scientific knowledge to provide simple explanations of everyday phenomena, evaluate basic aspects of scientific enquiry, and identify relevant information or evidence to support a conclusion, as well as justify its use. In this sense, Level 2 represents a shift from knowing to using science, even if at a basic level. It provides a meaningful reference point for assessing whether students can apply scientific knowledge to explain, interpret and evaluate straightforward situations. More specifically, at Level 2, students can use science to:

• Distinguish an appropriate scientific explanation from a non-scientific one.

• Give a simple scientific explanation of an everyday phenomenon.

• Evaluate simple enquiry designs.

• Interpret simple data relationships, including outliers.

• Identify evidence in a graph or table that supports a claim.

• Select a relevant information source for a science-related decision.

• Judge source credibility using a basic scientific criterion, such as expertise, scientific consensus or basic science.

Level 2 is the point at which students are beginning to use science as a tool for reasoning: explaining, interpreting, evaluating and making decisions in familiar real-world situations. The baseline is therefore meaningful because it represents a threshold of functional science literacy.

Students who do not attain baseline Level 2 are referred to in this report as “low performers”. Low-performing students are at greater risk of having weaker mastery of the foundational skills needed for further learning and meaningful participation in society, placing them at greater risk of adverse educational and labour-market outcomes later in life (OECD, 2016[2]; OECD, 2023[3]; OECD, 2025[4]).

As shown in Figure I.2.3, on average across OECD countries, 74% of students scored at Level 2 or above in science proficiency. In 16 countries and economies, at least 80% of students scored at Level 2 or above.

The proportion of top-performing students indicates how successful systems are in not only securing basic skills for the majority, but also in nurturing advanced proficiency. Some 7% of students attained the highest proficiency levels, Level 5 or 6, in science on average across OECD countries. In 14 countries/economies, the share of top performers in science was over 10%.

On average across OECD countries, 26% of students did not reach the baseline proficiency Level 2 in science (Table I.B1.2a.10). In seven countries and economies, less than 15% of students performed below baseline proficiency (B-S-J-Z (China), Macao (China), Singapore, Estonia, Japan, Chinese Taipei and Korea, in ascending order of the proportion of low performers). In these seven systems, most of the relatively few low-performing students scored at Level 1a, the highest proficiency level below baseline, meaning that few students performed far below the baseline in science. Over 75% of students performed below baseline proficiency in 11 systems; among them, over 10% of students performed below Level 1b in Dushanbe (Tajikistan), Kenya, Paraguay, Rwanda and Zambia. The share of students below Level 1b also exceeded 10% in Lebanon, although fewer than 75% of students in that system performed below baseline proficiency overall.

These results show that mean performance and distribution across proficiency levels provide complementary information about the quality and inclusiveness of education systems. A system may perform well on average because many students reach high levels of proficiency, because few students fall below baseline expectations, or both. Conversely, lower mean performance often coincides with a larger concentration of students below Level 2, indicating that weak overall performance is closely linked to difficulties in ensuring that a broad majority of students acquire foundational scientific competencies.

In PISA 2025, a novel assessment was included to test computational problem solving. It focuses on students’ capacity to engage in an iterative and self-regulated process of knowledge building and problem solving using computational tools and practices. Students complete two extended test units, during which they solve increasingly complex tasks and have access to learning resources such as tutorials, feedback and examples.

Students’ performance on the computational problem-solving scale describes the extent to which students can solve problems using computational tools, such as modelling or programming tools, and through computational practices, such as conducting experiments, analysing data, and building and debugging computational artefacts. Students’ performance on the test is supported by self-regulated learning, which includes how they maintain their motivation and task engagement, monitor their progress and adapt their strategies, and evaluate their own performance. A student's ability to succeed in the digital world is greatly influenced by their self-regulated learning competencies, i.e. how they manage their own learning, and, crucially, whether they can maintain their motivation and stay engaged with a task, even when it becomes challenging or frustrating.4

In computational problem solving, the average score among OECD countries is 500 score points (Table I.B1.2a.4). The highest-performing education systems are Macao (China), Singapore, B-S-J-Z (China), Japan, Chinese Taipei and Hong Kong (China) (in descending order and all above 545 score points). Australia, Estonia, Japan, Korea and New Zealand* are the top OECD countries (above 530 score points) (Table I.2.4). In addition to these ten countries and economies, another 15 education systems also performed above the OECD average in computational problem solving, ranging from Canada* (mean score of 525 points) to Luxembourg (mean score of 506 points).

Systems that consistently perform above the OECD average across assessment domains can be considered to demonstrate broad-based achievement. Consistent with the pattern observed earlier for science, mathematics and reading, almost all countries and economies that performed above the OECD average in computational problem-solving also performed above the OECD average in science, except for Denmark, Latvia, Luxembourg and the Netherlands*. Similarly, all countries and economies that performed above the OECD average in computational problem solving also performed above the OECD average in mathematics and reading, except for Latvia and Luxembourg, and except in reading only for Belgium, Denmark, Lithuania and the Netherlands*, and the United States* for mathematics only (Table I.2.1, I.2.o1 (online), I.2.o2 (online) and Table I.2.4).

Finally, to assess 15-year-old students’ computational problem-solving skills, the PISA 2025 assessment scale is divided into descriptive proficiency levels, ranging from Level 6 (the highest) down to below Level 1 (the lowest). The main assessment scale represents students’ overall computational problem-solving ability and reflects progress in both block-based programming and computer modelling. More specifically, it looks at whether students can carry out practical actions such as conducting systematic experiments, analysing the resulting data, and building or debugging their digital creations (the full computational problem-solving proficiency descriptors for PISA 2025 are included in Table I.2.o10 (online)).

Students performing below Level 1 are still largely exploring the software interface in a rudimentary way, rather than working towards a specific goal. Once they reach Level 1, they begin to interact with these tools at a surface level. They may make minor modifications to simple programs or adjust a single input in a basic model to observe what happens. This is still a relatively elementary form of manipulation of the digital environment.

At Levels 2 and 3, students begin to make more meaningful progress. At Level 2, they start to use simple operations, such as instructing a program to repeat an action, and begin testing out ideas. By Level 3, they combine a wider range of commands to build functional programs. At this level, students also begin to conduct basic, controlled experiments in order to infer simple relationships between different factors in a model. Although there is no formal baseline proficiency threshold on this scale, Level 3 can be regarded as a useful benchmark, as it marks the point at which students begin to demonstrate sufficiently solid knowledge and skills to progress more rapidly.

At Level 4, students are able to build moderately complex programs and models. They begin to exploit repeated patterns in their code, even if their solutions may still contain minor bugs or inefficiencies. Levels 5 and 6 can be considered the range of top performers. By Level 5, programming and modelling are relatively sophisticated, while Level 6 represents highly developed proficiency. Students at this level are efficient computational problem solvers who can produce optimal, problem-free programs without unnecessary redundancy.

Across OECD countries, about one-quarter of students can be considered top performers in computational problem solving (26%). A further 21% reach Level 4 and 17% reach Level 3, while 13% perform at Level 2 and 17% at Level 1. Only 5% perform below Level 1 (Table I.B1.2a.13).

To what extent does the PISA computational problem-solving assessment measure a different set of skills with respect to those measured in the core assessment domains?

In a context where the use of artificial intelligence is on the rise, solving problems using computational tools and practices is a fundamental skill. Across education systems, these practices are increasingly embedded within “traditional” academic domains. Because the process of building computational artefacts naturally lends itself to applying core practices and concepts from broader STEM disciplines, computational thinking is often integrated into science curricula. This highlights a strong, natural synergy between scientific inquiry and computational problem-solving skills (OECD, 2023[5]).

It can therefore be expected that student performance in computational problem-solving correlates positively with performance in other academic fields, even if the PISA 2025 computational problem-solving assessment places more emphasis on how students learn to use computational tools and apply computational practices in interactive, digital environments (OECD, 2023[5])). Students who perform well in computational problem-solving are likely to perform well in other subject areas, just as students who do not achieve high scores in science, mathematics and reading are likely to achieve low scores in computational problem-solving. At the same time, the assessment is designed to capture a set of competencies that is related to, but not reducible to, performance in other PISA domains. This section examines these associations in detail, including the relative performance of students in computational problem-solving given their performance in science, mathematics and reading.

Table I.2.5 shows the OECD average within-country correlations between student performance in computational problem-solving, mathematics, reading and science. The within-country correlation between computational problem-solving and each of the other three PISA domains is: 0.76 with science, 0.69 with reading and 0.75 with mathematics. In comparison, performance scores among the three core PISA domains are more strongly associated, especially between performance in science and mathematics (0.88). This suggests that while strongly correlated, the computational problem-solving performance measures a distinct set of skills relative to those measured in the science, reading and mathematics performance (Table I.B1.2a.19). One distinctive feature of the computational problem-solving assessment is that students can make use of feedback and resources during the test. This means that performance in this domain reflects not only what students already know, but also how effectively they can learn, adapt and respond within the assessment itself. In this respect, the interpretation of performance differs somewhat from that in the other three domains.

Another way of evaluating the uniqueness of the skills measured by the computational problem-solving scale is to examine the variation in student performance in computational problem-solving that cannot be associated with their performance in another scale. On average across the OECD, around 43% of the variation in computational problem-solving performance can be uniquely associated with student performance in science (the major domain focus in PISA 2025) (Table I.B1.2a.19). This means that an important part of the variation in performance can be accounted for simply by student performance in science. At the same time, more than half of the variation in computational problem-solving remains outside its unique association with science5, which further supports the view that computational problem-solving captures additional competencies beyond those measured in another domain such as science.

In 11 countries and economies (Albania*, Armenia, Azerbaijan, Bulgaria, Georgia, Hungary, Kyrgyzstan, Lebanon, Morocco, the Palestinian Authority and Uzbekistan), less than 30% of the variation in students’ computational problem-solving performance is associated with their performance in science uniquely. In these countries and economies, more than in others, performance differences in computational problem-solving do not necessarily match those found in science. For example, some students who perform at a high level of proficiency in science achieve relatively low scores in computational problem-solving (and vice versa) in these countries/economies. By contrast, in Iceland, Brunei Darussalam, Qatar, New Zealand*, Australia, Ireland, Estonia and Canada* (in ascending order), around 50% or more of the variation in computational problem-solving performance reflects performance differences captured uniquely in the science assessment (Table I.B1.2a.19).

In some countries and economies, students had relatively higher or lower scores than expected in computational problem-solving given their scores on the other PISA assessments. In this section, the statistical concept "relative performance"6 is used to examine how well students performed in computational problem solving compared to other students who have similar science, mathematics or reading scores.

For example, if a country or economy has a positive relative score, it means that its students performed better in computational problem solving than other students with similar science proficiency in the PISA-participating countries and economies. Figure I.2.4 shows the share of students who performed above and below what would be expected given their performance in science, reading or mathematics. In Brunei Darussalam, Hong Kong (China), Macao (China), Malaysia and the Philippines, , over 70% of students scored significantly higher than expected in computational problem solving given their performance in science (Table I.B1.2a.20).

In six countries/economies (Hong Kong (China), Malaysia, Brunei Darussalam, the Philippines, Macao (China) and Iceland, in descending order), students scored over 30 points higher than expected in computational problem solving – a large relative performance advantage – after accounting for their science performance7 (Table I.B1.2a.20). These six countries/economies include education systems across different levels of the computational problem-solving performance distribution (mean scores within proficiency Levels 2 to 5), including both high- and mid-performing systems (but no low-performing ones). This suggests that relatively strong performance in computational problem-solving is not confined to the very highest-performing systems and may also be observed among systems where students display only average levels of science proficiency.

By some margin, the country with the weakest relative performance after accounting for students’ science performance is Kenya, with students scoring about 87 points lower in computational problem-solving than expected, followed by Uzbekistan, Azerbaijan and Armenia (55, 43 and 42 points lower, respectively). All countries with a negative relative performance in computational problem-solving after accounting for their science performance scored below the OECD average, except for Finland and top performer B-S-J-Z (China) (Table I.B1.2a.20).

PISA data for 2022 and 2025 show contrasting patterns of mean performance across science, reading and mathematics. On average across 35 OECD countries, mean performance did not change significantly in science between 2022 and 2025. By contrast, mean performance dropped by 9 score points in mathematics and about 14 score points in reading (Tables I.B1.2a.36, I.B1.2a.38 and I.B1.2a.37).

As shown in Table I.2.6 , science performance remained broadly stable in many countries/economies (48 out of 74) between 2022 and 2025. In contrast, in reading and mathematics, at least half of countries/economies that can compare PISA 2025 and 2022 results showed a drop in mean performance (42 and 38 out of 74 in reading and mathematics, respectively).

In many cases, the drop observed over just three years exceeded 20 score points, i.e. the yearly gain in test scores that is typically observed among students around the age of 15 (see Box I.2.2). This means that 15-year-olds in these countries in 2025 scored at or below the level expected of 14-year-olds in 2022.

• In science, the decline in performance exceeded 20 score points in Guatemala, Hong Kong (China), Israel and Latvia.

• In mathematics, the decline in performance was most pronounced and exceeded 20 score points in Israel, Latvia, Malta, Serbia and Slovenia.

• In reading, the decline in performance exceeded 30 score points in Latvia, Israel, Guatemala and Malta (in descending order). Drops between 20 and 30 score points were observed in Croatia, Czechia, Denmark, Hungary, Norway*, Paraguay, Serbia, Slovenia, Spain and Sweden.

Many more countries and economies than those listed above experienced performance declines between 2022 and 2025.

However, as shown in Table I.2.6, it is equally relevant to note that eight countries and economies improved their performance in all three subjects. Systems showing positive trends in at least one domain do not form a homogeneous group. Some are at the upper end of the average performance range in at least one domain and have sustained or improved upon already strong outcomes (e.g. Türkiye and the United Kingdom,), others are lower-performing systems that appear to be catching up (e.g. Cambodia, El Salvador, Jordan and the Philippines), sometimes by expanding access to effective schooling. Positive trends should therefore be interpreted in light of systems’ starting points, mean levels of performance, and the distribution of achievement within systems. Several of these characteristics will be analysed later on in this chapter.

The change in mean science, mathematics and reading performance across OECD countries (on average) and in most PISA-participating education systems was not uniform across low- and high-achieving students. While the average mean performance in science did not change significantly – contrasting sharply with the drops in reading and mathematics – looking only at the stable average ignores widening performance gaps. The performance gap between the highest- and lowest-achieving students in science actually widened by about 3 score points between 2022 and 2025 on average across 35 OECD countries, and by more than 15 score points between 2018 and 2025 (Table I.B1.2a.42). Indeed, PISA 2025 data suggest that scientific competencies have diminished among students on the lower end of the performance distribution, and similar issues affecting reading and mathematics could be affecting low-achieving students in science.

The 10th percentile is the point on the scale below which only 10% of students score. Likewise, the 90th percentile is the point on the scale above which only 10% of students score. The median or 50th percentile divides the performance distribution into two equal halves, one above and one below that position on the scale.

Across OECD countries, performance gaps had widened in the past, and continued to do so between 2022 and 2025. The mean scores of the lowest-performing 10% of students are the lowest reported since PISA started measuring science in 2006, on average across 35 OECD countries with comparable data across cycles. At the other end of the distribution, the mean scores of the highest-performing 10% of students have remained stable since PISA started assessing science (Table I.B1.2a.39). As a result, the gap between the highest- and lowest-performing students continued to widen; at 259 score points, it is the largest measured in PISA so far, on average, when measured by the inter-decile range (i.e. the difference between the 10th and 90th percentiles) (Table I.B1.2a.42).

In mathematics, mean performance was about 9 score points lower in 2025 compared to 2022 on average across 35 OECD countries. But the performance decline was less pronounced at the 75th and 90th percentiles (-7 and -5 score points): almost all students performed worse, but low-achieving students declined more than high-achieving students did (Table I.B1.2a.41). A similar pattern is observed in reading: the decline was slightly less pronounced at the 90th percentile (-13 score points), even if the drop was large across the score distribution (Table I.B1.2a.40). As a result, learning gaps between the highest- and lowest-performing students widened (see Chapter I, Figure I.2.6). The difference between the 10th and 90th percentiles increased by about 3 score points in reading between 2022 and 2025, but by about 8 score points in mathematics, on average across OECD countries (Tables I.B1.2a.43 and I.B1.2a.44).

The previous paragraphs refer to the average trend across 35 OECD countries; however, the distribution in performance evolved differently in different countries and economies. For example, in science, the inter-decile range widened significantly in 24 countries and economies; narrowed significantly in one (Peru); and did not change significantly in the remaining countries/economies for which comparable data for 2022 and 2025 are available (Table I.B1.2a.42).

In mathematics, the inter-decile range widened in 20 countries and economies and narrowed only in three, but did not change significantly in most countries/economies (Table I.B1.2a.44). Finally, in reading, the inter-decile range did not change significantly in most countries/economies. It widened in 10 and narrowed in 11 (Table I.B1.2a.43).

The performance distribution widened between 2022 and 2025 in all three subjects in Singapore, North Macedonia, Japan and Latvia, as well as on average across OECD countries (Tables I.B1.2a.42, I.B1.2a.44and I.B1.2a.43).

PISA scores in science, reading and mathematics are more than a tool to rank students and countries. Together with descriptive proficiency scales, scores give information on the level of science competencies students have mastered (see section above on What students can do in science). In science, at the highest proficiency levels, students can draw on high cognitive-demand scientific ideas to build models of complex phenomena, evaluate competing designs for scientific enquiries, and use sophisticated fact-checking procedures to critique the trustworthiness of scientific information. At the lower levels, students can identify a scientific explanation of a simple phenomenon drawing on basic scientific knowledge, information or evidence of low cognitive demand, and use basic procedural knowledge to identify the better interpretation or display of simple data (see Table I.2.2).

Trends in the proportion of low- and top-performing students indicate how their mastery of specific skills (as established in the described proficiency scale) has changed over time. Changes in the share of low-performing students indicate the extent to which school systems are advancing (or not) towards providing all students with basic skills. Trends in the share of top-performing students indicate whether education systems are making progress in ensuring that young people can successfully use their science, mathematics and reading skills to navigate a volatile, uncertain and complex environment.

As shown in Table I.2.8, on average across the 35 OECD countries with comparable data, the shares of both low- and top-performing students in science did not change significantly between 2022 and 2025. This stability at the OECD average masks some variation across systems. The share of top performers in science did not change significantly in the majority of systems (61), increased in 7 systems and decreased in 7 systems. Similarly, the share of low performers remained unchanged in many systems (45), increased in 14 systems and decreased in 16 systems. Only in Denmark and Latvia, did the shares of low performers increase and those of top performers decrease. By contrast, in Jordan, Uzbekistan, Montenegro, Saudi Arabia, Türkiye, the United Arab Emirates and the United Kingdom, the share of low performers decreased while the share of top performers increased,

Short-term changes between 2022 and 2025 provide only a partial view of how performance distributions are evolving. A decade-long perspective shows whether recent stability reflects genuine continuity or instead conceals gradual shifts in the shares of students below baseline proficiency and at the highest proficiency levels.

On average across the 35 OECD countries with available data, the proportion of students scoring below Level 2 in science (low-performing students) increased by three percentage points between 2015 and 2025, whereas the proportion of students scoring at or above Level 5 (top-performing students) remained unchanged over this period (Table I.B1.2a.33). Over the decade prior to 2025, 13 countries and economies showed a similar pattern of increasing shares of low-performing students and unchanged shares of top-performing students. In 8 countries and economies, the share of low-performing students increased while the share of top-performing students decreased (Table I.B1.2a.33).

The proportion of students who do not reach Level 2 on the PISA scales and the proportion of students who are able to score at Level 5 or 6, together indicate the characteristics of a country’s/economy’s talent pool. Only five countries/economies, Montenegro, Qatar, the Slovak Republic, Türkiye and the United Arab Emirates, , were able to simultaneously reduce their shares of low-performing students and increase their shares of top-performing students over the past decade. Twelve countries/economies reduced their share of low-performing students. In addition, Colombia, Korea, Macao (China), Chinese Taipei and the United States*, increased their share of students at Level 5 or 6 (Table I.B1.2a.33).

In reading, the proportion of students scoring below Level 2 (low-performing students) increased by 10 percentage points between 2015 and 2025, on average, whereas the proportion of students scoring at Level 5 or 6 (top-performing students) decreased by two percentage points (Table I.B1.2a.34). A similar change was also observed in mathematics, with an increase of about 9 percentage points in the proportion of students scoring below Level 2 and a decrease of about two percentage points in the share of top-performing students (Table I.B1.2a.35).

In 25 years of PISA, since the first PISA assessment in 2000, the average trend across OECD countries is negative (Figure I.2.6). Performance in PISA 2025 was the lowest in all subjects, significantly below the mean performance observed in any earlier assessment (except for science in PISA 2022). In science and reading, the strongest performance was observed in 2009 and 2012, respectively, then the trajectory turned negative. In mathematics, performance remained close to the 2003 level through all assessments up to 2018, then dropped sharply between 2018 and 2025.

Over the most recent decade (2015-2025), the trend in performance has been negative in all three core subjects, on average across 35 OECD countries with available data. Indeed, the recent stability in science performance masks a sustained, long-term decline. On average across OECD countries, the decennial trend is negative, indicating an average drop of 7 score points in science performance since 2015 (Table I.B1.2a.36).

As shown in Table I.2.9, such decade-long negative trends can also be seen in 28 countries and economies, with average negative trends ranging from 7 score points in Malta to 37 in Iceland, over the same period. However, in a number of countries and economies, PISA data show average decade-long positive trends in science, ranging from an average trend of 7 score points in Singapore to 62 in Türkiye and 72 in Cambodia.

Figure I.2.7 shows the linear trend in median performance since PISA 2015, alongside trends observed in the performance of students in the 10th and 90th percentiles.

Trends at the 10th percentile indicate whether the lowest-achieving 10% of students in a country/economy moved up the PISA scale over time. Similarly, trends at the 90th percentile indicate improvements among a country’s/economy’s high-achieving students.

Among countries and economies whose mean science performance worsened since 2015, there have been both widening and shrinking performance gaps:

• In Denmark, Finland, Iceland, North Macedonia, the Netherlands*, Norway*, Portugal and Slovenia, declines among both low-achieving and high-achieving students resulted in the widening in the achievement gap in science (measured by the distance between the 10th and 90th percentiles) between 2015 and 2025.

• In Belgium, Greece, Indonesia and Israel, there were declines among both low-achieving and high-achieving students, but the gap remained unchanged.

• Only in Kosovo did more rapid declines among high-achieving students result in the achievement gap in science shrinking between 2015 and 2025.

• In Canada*, Germany, Hong Kong (China), Latvia, Lebanon, Poland and Spain, as was the case on average across the 35 OECD countries that can compare PISA results across all assessments, the long-term decline occurred among low-achieving students only, resulting in a widening gap with respect to high-achieving students whose scores did not change over the decade (Tables I.B1.2a.36, I.B1.2a.39 and I.B1.2a.42).

Among countries and economies where science performance improved over the 2015-2025 period, Lithuania, Montenegro and Qatar saw no significant change in the achievement gap, although low-achieving students’ results improved. In Korea, Macao (China), Chinese Taipei and the United Arab Emirates, the average improvement was due to high-achieving students improving while low-achieving students showed no significant change, which widened the performance gap. Only in Georgia did performance in science among low-achieving students improve while high‑performing students remained unchanged, resulting in a shrinking performance gap (Tables I.B1.2a.36, I.B1.2a.39and I.B1.2a.42).

Finally, among countries and economies with no significant change in mean science performance over the 2015‑2025 period, Austria, Brazil, Colombia, Costa Rica, Croatia, Ireland, Japan, Romania, Sweden, the United Kingdom and the United States* nevertheless experienced a widening of their performance gap.

On average across the 35 OECD countries that can compare PISA results across all assessments, performance differences widened in reading and mathematics even though both low-achieving students and high-achieving students performed worse; however, low-achieving students declined by more than high-achieving students did (Tables I.B1.2a.37, I.B1.2a.40, I.B1.2a.43, I.B1.2a.38, I.B1.2a.41 and I.B1.2a.44).

When looking at reading performance, in countries/economies with declining decade-long trends, the drop was driven by low-achieving students (i.e. changes were statistically significant only for low-achieving students) in Australia, Costa Rica, Korea, Macao (China), Mexico, New Zealand*, Romania, Switzerland, and Uruguay. In these systems, the result was a widening of the performance gap, with the exceptions of Mexico, Romania and Uruguay. At the opposite end, the average drop was driven by a drop in performance among high-achieving students in Georgia, Lebanon, Matla, Moldova and Montenegro, which resulted in a shrinking of the performance gap.

In countries/economies with declining decade-long trends in mathematics performance, the drop was driven by low‑achieving students in Australia, Czechia, Estonia, Hong Kong (China), the Netherlands*, New Zealand*, Romania, the Slovak Republic, Switzerland and the United Kingdom. In all of these systems, the result was a widening of the performance gap. Moreover, the average drop was driven by a drop in performance among high-achieving students, in Chile, Colombia, Indonesia, Kosovo, Lebanon, Mexico, Moldova, Montenegro and Thailand, . In these systems, this resulted in a shrinking of the performance gap with the exceptions of Colombia and Thailand, where the gap remained unchanged.

In most countries and economies, all boys and girls who were born in 2009 were of the correct age to sit the PISA 2025 test (in countries and economies that tested students during the second part of 2025, a 12-month period spanning the years 2009 and 2010 defined the eligible birthdates). However, age was not the only criterion for eligibility: 15-year-olds also had to be enrolled in 7th Grade or higher at the time of testing.

Eligibility in PISA is determined by more than just a student’s age, the PISA sample excludes 15-year-olds who do not go to school or who are severely delayed in their school-grade progression. However, PISA results still reflect a combination of 15-year-olds’ access to education and the quality of the education they have received over the course of their lives.

Globally, enrolment in secondary education has continued to expand over the past decade in many countries. As a result, PISA coverage – the proportion obtained by dividing the number of PISA-eligible students by the total number of 15-year-olds in a country – has increased in a number of participating countries and economies.

Many of these countries and economies saw a declining trend over the past decade in one or two and, sometimes, in all three subjects. These declines could mean that the quality of education has gone down for all students in the past decade, or they could also reflect the expansion of education to more marginalised populations.

The expansion of education systems is also reflected in PISA data: in all of the 12 countries where fewer than two in three 15-year-olds were eligible to participate in past PISA assessments, there is now an increased share of 15‑year‑olds eligible for the test relative to all the 15-year-olds in the country. Most notably, over the past decade, Cambodia, Ecuador, El Salvador, Lebanon, Morocco and Paraguay saw increases between 11 and 24 percentage points in the PISA coverage – the proportion obtained by dividing the number of PISA-eligible students by the total number of 15-year-olds in a country8.

However, this expansion in education opportunities makes it more difficult to interpret how mean scores in PISA have changed over time. Increases in the share of PISA-eligible students relative to all 15-year-olds can lead to an underestimation of the real improvements education systems have achieved. Household surveys often show that children from poor households, rural areas or ethnic minorities have a greater chance of not attending or completing lower secondary education (GEM Report UNESCO, 2015[21]) Typically, as populations that had previously been excluded gain access to higher levels of schooling, a larger proportion of low-performing students would be included in PISA samples (Avvisati, 2017[22]).

Of the countries in which PISA coverage increased markedly since 2015, only Cambodia, the Dominican Republic and El Salvador saw improvements in performance in at least one subject by 2025. Among those with a longer record of participation, mean performance in science remained stable in Colombia, Costa Rica and Mexico, although it decreased in reading or mathematics (or both) over this period (Tables I.B1.2a.36, I.B1.2a.37 and I.B1.2a.38). Only in Brazil, performance remained stable in the three domains since 2015, while coverage increased by four percentage points.

Do declines mean that the quality of education has gone down for all students in the past decade, or do they reflect the expansion of education to more marginalised populations? By considering a population equal in size to 25% of an age group made up of only the best-performing students in a country, it is possible to monitor the rate of change in PISA performance for a sample of 15-year-olds that was only indirectly affected by changes in coverage rates over a given period, but whose composition remained unchanged. Most likely, all members of this group would have been eligible to participate in PISA even in the counterfactual situation of no educational expansion. This analysis offers a different interpretation of many of these countries’ results.

In Cambodia, minimum scores for 25% of top-performing 15-year-olds increased over this period in all three subjects. In Paraguay, minimum scores for 25% of top-performing 15-year-olds increased in science and mathematics and remained stable in reading. In Costa Rica, minimum scores for 25% of top-performing 15-year-olds increased over this period in science and remained stable in reading and mathematics. Moreover, in Guatemala and Morocco, minimum scores for 25% of top-performing 15-year-olds remain unchanged in at least two domains, and only in one in Mexico (Tables I.B1.2a.48, I.B1.2a.49 and I.B1.2a.50).

In sum, among the seven countries that increased participation in secondary education over the 2015-2025 period and for which performance change data are available, mean scores improved in all subjects in Cambodia, in two subjects in the Dominican Republic and in one in El Salvador; dropped in all subjects in Guatemala, and in at least one subject in the remaining four (Costa Rica, Mexico, Morocco and Paraguay). In these systems, the decline in mean scores is likely linked to the integration of more 15-year-olds from marginalised populations into schooling. PISA results show that in at least three of these education systems, performance did not deteriorate and that expanding secondary education to more marginalised students was not necessarily accompanied by lower performance among their higher-performing peers.

The analyses above show that changes in average performance often conceal important shifts in the distribution of achievement, especially at the lower end. At the same time, mean performance remains an important part of equity in education: an education system cannot be considered fully equitable if most students perform at low levels, even when differences between social groups are small. To understand whether differences in outcomes reflect inequalities in learning opportunities, it is therefore necessary to examine how performance is associated with students’ social and educational backgrounds, while keeping the overall levels of achievement in view.

Equity is a fundamental value and a central goal of education policy. From a social justice perspective, it reflects the principle that all individuals, regardless of their background, should have the opportunity to fulfil their potential. From an economic and societal perspective, it also matters because education systems are stronger when they enable all learners to develop their capabilities, rather than allowing talent and potential to go unrealised. From a global policy perspective, it is also embedded in the 2030 Agenda for Sustainable Development, particularly in Sustainable Development Goal 4, which calls for “inclusive and equitable quality education” for all, and in Target 4.5, which aims to eliminate disparities in education and ensure equal access for vulnerable groups. In addition, Target 4.1 underscores the importance not only of access, but also of ensuring that all girls and boys equitably achieve relevant and effective learning outcomes (United Nations, 2015).

Yet achieving genuine equity remains one of the most complex challenges facing school systems today. Although global expansions in schooling over the past century have dramatically increased access to education, socio-economic inequalities in learning outcomes have proven remarkably persistent (Chmielewski, 2019[25]; Pfeffer, 2008[26]). Equity is understood, first of all, as an ethical imperative to ensure that personal or social circumstances – factors over which students have no control – do not dictate their opportunity to fulfil their academic potential (OECD, 2023[3]). But equity also requires that this potential be realised at meaningful levels of learning: reducing disparities is not sufficient if students from all backgrounds are left with weak skills.

This section analyses two key dimensions of equity in education: inclusion and fairness. Only education systems that combine high levels of both are considered highly equitable. Fairness refers to the extent to which all students are given the opportunity to realise their full learning potential, irrespective of their background. Inclusion, in PISA, refers to whether all young people have access to quality education and acquire at least the baseline skills needed to continue learning to achieve their potential and fulfil their aspirations. However, baseline proficiency should be understood as a minimum threshold, not as the endpoint of equity. This is why the equity indicators discussed in this section need to be read alongside the earlier results on mean performance: an education system in which most students reach only baseline proficiency, or in which performance is uniformly low regardless of background, would not represent the full ambition of equity. Equity requires both broad access to foundational skills and meaningful opportunities for students from all backgrounds to progress towards higher levels of achievement.

In PISA 2025, only education systems that combine high levels of both inclusion and fairness are considered equitable. Inclusion is measured by the percentage of 15-year-olds who are at or above proficiency Level 2 in at least one subject (as will be discussed, this reflects the assumption that 15-year-olds not covered by the PISA sample would score below Level 2).9 Fairness is measured by the percentage of variance in science performance accounted for by student socio-economic status. Figure I.2.10 shows countries and economies according to their levels of inclusion and fairness in science.

In 18 out of the 28 countries and economies that had a level of inclusion above the OECD average (i.e. more than 65% of 15-year-olds scored at or above proficiency Level 2 in science), the level of fairness by socio-economic status was also significantly above the OECD average (i.e. less than 12% of the variance in science performance was accounted for by student socio-economic status). Education systems in B-S-J-Z (China), Canada*, Estonia, Ireland, Japan, Korea, Lithuania, and Macao (China) achieved high levels of both inclusion (over 75%) and fairness (under 10% variance). Importantly, these systems did so while also attaining strong overall outcomes: the average score in science, reading and mathematics was higher than the OECD average (except in Lithuania, where the mean score in reading was not statistically significantly different from the OECD average). While these results do not imply that equity causes higher performance, they indicate that many systems that are most successful in limiting low performance and reducing socio-economic disparities also tend to be systems where overall achievement is strong. They can therefore be considered highly equitable (Tables I.B1.2a.45 and I.B1.2b.2).

These results show that high levels of equity are not incompatible with strong overall performance; on the contrary, some education systems demonstrate that it is possible to combine broad inclusion, weak socio-economic disparities and high achievement across subjects. This distinction is important. An education system in which most students reach only baseline proficiency, or in which performance is uniformly low regardless of background, would not represent the full ambition of equity. Equity requires that students from all backgrounds have access to foundational skills and meaningful opportunities to advance towards higher levels of achievement.

In PISA, educational inclusion is assessed through two complementary measures: the extent to which all young people secure access to schooling – i.e. the share of 15-year-olds who are enrolled in school and represented in the assessment – and the extent to which students attain at least a baseline level of proficiency in core subjects such as science, reading and mathematics by the age of 15 (OECD, 2016[2]). This measure combines aspects related to the “quantity” of schooling (i.e. the share of 15-year-olds who are enrolled in school) with measures of the “quality” of education outcomes (i.e. the share of students who scored at least at the minimum level of proficiency).

Inclusion is particularly important for students from socio-economically disadvantaged or otherwise marginalised backgrounds. These students often face multiple risk factors that make them especially vulnerable to low proficiency levels and early school withdrawal (De Witte et al., 2013[27]; OECD, 2016[2]). In practice, a truly inclusive education system ensures that the vast majority of young people are not only enrolled in school, but also acquire essential baseline cognitive skills.

Expanding school participation without improving learning outcomes is not sufficient for economic and social success. Comprehensive assessments of educational inclusion must therefore consider both school enrolment and the acquisition of essential cognitive skills, so that young people are not left behind.

Baseline proficiency should therefore be understood as a minimum threshold, not as the endpoint of equity. An education system that enables most students to reach Level 2, but does not provide equal opportunities for students from all backgrounds to progress further, cannot be considered fully equitable. Equity requires both broad access to foundational skills and the opportunity for all students to develop their potential beyond the baseline.

Access to schooling does not in itself guarantee that students will acquire the knowledge and skills needed to participate fully in contemporary societies, but it is, of course, a necessary precondition for educational inclusion. Access is reflected most directly in the extent to which 15-year-olds are still enrolled in school, while early school leaving and marked delays in grade progression signal weaker inclusion.

These indicators matter because students who leave school early are less likely to fully benefit from the educational, social and labour-market advantages associated with continued participation in education, and because school dropout is shaped by a combination of individual, family, school and institutional factors (Spaull and Taylor, 2015[28]; Hanushek and Woessmann, 2008[29]). Education systems with broader school coverage at age 15 and lower levels of early school leaving or severe delay can therefore be regarded as more inclusive.

While PISA is not designed to estimate enrolment rates directly, it provides a range of indices that capture the coverage of the population of 15-year-olds enrolled in Grade 7 or above in each country and economy. Low values of the proportion of 15-year-olds in each country/economy covered by the PISA sample (Coverage Index 3) may reflect 15-year-olds who are no longer enrolled in school or who have been held back in primary school. This index may also be reduced by student exclusions from the PISA assessment and by dropout during the school year.

Coverage Index 3 ranges from 32% in Rwanda and 43% in Cambodia to 90% or more in 34 countries and economies (see Annex A2). While PISA results are representative of the target population in all adjudicated countries and economies, they cannot be readily generalised to the entire population of 15-year-olds in countries where many young people of that age are not enrolled in lower or upper secondary school.

Coverage is only one component of inclusion. The next question is whether students who remain in school also acquire the minimum competencies needed to participate fully in society.

In this chapter, low performance has been examined separately in science, mathematics and reading. Yet low achievement is often cumulative: students who do not attain baseline proficiency in one domain are at heightened risk of falling below baseline in others. A fuller account of low performance therefore requires attention not only to subject-specific rates, but also to the overlap in low performance across domains, especially the share of students who fail to reach baseline proficiency in all three subjects (OECD, 2023[3]).

At the same time, the results presented so far describe only the 15-year-olds covered by the PISA 2025 target population and sample. In some participating countries and economies, however, a non-trivial share of 15-year-olds was not represented in the assessment. Incomplete coverage can affect how national skill distributions are interpreted, particularly where school attendance at age 15 is not universal.

Because the achievement of 15-year-olds not covered by PISA is not directly observed, their likely position in the skills distribution must be estimated indirectly. Evidence on PISA sample eligibility suggests that students excluded from the sampling frame are often out of school or severely over-age for grade, and that ignoring these groups can lead to an underestimation of both the prevalence of low skills and the extent of inequality in the broader youth population (Spaull, 2017[30]).

More broadly, the literature shows that educational exclusion and early school leaving are associated with clusters of disadvantage rather than a single risk factor, and that school dropout reflects interacting individual, family, school and system-level processes (De Witte et al., 2013[27]; Gubbels, van der Put and Assink, 2019[31]). In low- and middle-income settings, combining data on educational attainment with data on cognitive outcomes further suggests that limited access and weak learning outcomes frequently accumulate among the same groups of young people (Spaull and Taylor, 2015[28]).

Rather than attributing an exact score to these 15-year-olds, it is possible to estimate lower and upper bounds for most results of interest, including the mean score, the median score and other percentiles, or the proportion of 15‑year-olds reaching minimum levels of proficiency (Avvisati, 2017[22]). Under a best-case scenario, where the distribution of reading, mathematics and science skills among the population not covered by the sample is assumed to be the same as that of the covered population, the estimates of mean scores and percentiles derived from PISA samples represent an upper bound on the mean, percentiles and proportions of students reaching minimum proficiency in the entire population of 15-year-olds. A lower bound can be estimated by assuming a plausible worst-case scenario, such as that all 15-year-olds not covered by the sample would score below a certain point in the distribution. For example, if all of these 15-year-olds were assumed to score below Level 2, then the lower bound on the proportion of 15-year-olds reaching minimum levels of proficiency would simply be the corresponding proportion observed in the PISA target population multiplied by Coverage Index 3.

Figure I.2.11 presents the proportion of 15-year-olds reaching minimum levels of proficiency under the assumption that all 15-year-olds not covered by the PISA sample would score below Level 2 in each subject.10 In the figure, 15‑year-olds are grouped according to whether they scored below the baseline level of proficiency in one subject only, in two subjects, or in all three core PISA subjects (science, reading and mathematics), as well as those not covered by the PISA sample, who are assumed to be low performers in all three subjects. The figure shows that all countries and economies that participated in PISA 2025 – even those with the highest levels of performance and equity – have a sizeable share of low performers.

The largest category of low-performing students is the group of 15-year-olds who scored below the baseline level of proficiency in all three subjects. On average across OECD countries, over one out of four students (29%) are low performers in science, reading and mathematics. This percentage includes both students who took the PISA test and 15-year-olds not covered by PISA, who account for 12% of the age cohort on average across OECD countries. In 18 countries and economies, more than 60% of 15-year-olds are low performers in all three subjects.

About 1% of students across OECD countries are low performers in science only; 4% are low performers in reading only; and 6% are low performers in mathematics only. About 3% are low performers in science and mathematics but not in reading; 2% are low performers in science and reading but not in mathematics; and 4% are low performers in reading and mathematics but not in science (Table I.B1.2a.24).

Taken together, all the categories of low performers shown in Figure I.2.11 represent the share of 15-year-olds who are low performers in at least one subject, whether mathematics, reading or science, plus those outside of the PISA target population. On average across OECD countries, 49% of 15-year-olds are low performers in at least one subject, although this share varies considerably across countries. In 47 countries and economies, more than 60% of 15-year-olds scored below baseline proficiency Level 2 in at least one subject. By contrast, in three countries and economies, fewer than 25% of 15-year-olds were low performers in at least one subject (Table I.B1.2a.22).

These results underscore that educational inclusion must be judged not only by how many students remain in school, but also by how many young people acquire at least the minimum competencies needed to participate fully in society.

A student's socio-economic status remains one of the most powerful and consistent predictors of academic achievement globally. However, the mechanisms through which family background translates into educational advantage are highly complex and multidimensional (Eriksson et al., 2021[32]). The roots of these disparities often take hold long before a child enters a formal school classroom. Research suggests that economic deprivation and adverse home conditions in early childhood can significantly hinder cognitive development and verbal ability, setting a trajectory of disadvantage that continues into adolescence and adulthood (Duncan, Brooks-Gunn and Klebanov, 1994[33]; Richards and Wadsworth, 2004[34]).

As students progress through school, disparities in resources can impact their academic progress. Socio‑economically advantaged students typically benefit from greater financial, social and cultural backgrounds (Paino and Renzulli, 2012[35]; Kao and Thompson, 2003[36]; Eriksson et al., 2021[32]; Coleman, 1988[37]; Bourdieu, 2018[38]). This makes it easier for them to succeed in school compared to students from families with lower levels of education or that are affected by chronic unemployment, low-paid jobs or poverty. Consequently, by the time students take the PISA test at age 15, a distinct socio-economic gradient can be observed, a measurable relationship demonstrating how deeply family background relates to educational outcomes within a given school system (Willms, 2006[39]).

However, socio-economic disadvantage does not determine academic success. Some disadvantaged students perform well despite adversity, demonstrating remarkable academic resilience (“resilient students”) and performing at the highest levels both nationally and internationally (OECD, 2011[40]). This section first examines socio-economic disparities in recent PISA results through the slope and strength of the socio-economic gradient across countries and economies, before looking at low performance and academic resilience.

PISA-participating countries and economies vary markedly in their levels of wealth and per-capita income (see Table I.B3.2.1), translating into differences in the socio-economic status of the students who take the PISA test. The average socio-economic status of students in each country and economy that participated in PISA 2025, is measured by the PISA index of economic, social and cultural status (ESCS) (see Annex A3 for a detailed definition of this index).

In PISA, the socio-economic gradient is used to examine the relationship between students’ socio-economic status and student performance in each country and economy whereby a stronger association means less fairness (thus, less equity) (Willms, 2006[39]). The socio-economic gradient offers two key pieces of information: the strength and the slope of the gradient.

The strength of the gradient is measured by the proportion of the variation in student performance that is accounted for by differences in student socio-economic status. When the relationship between socio-economic status and performance is strong, socio-economic status is a good predictor of performance. Meanwhile, the slope of the socio-economic gradient indicates the degree of the disparity in average performance between two students whose socio-economic status differs by one unit in the PISA ESCS index. A positive value for the slope of the socio-economic gradient signals that advantaged students generally performed better than disadvantaged students.

On average across OECD countries in 2025, students’ socio-economic status accounts for a significant share of the variation in their performance in PISA. PISA data show that 12% of the variation in science performance within each country is associated with socio-economic status, on average. In Ecuador, Luxembourg, Romania, the Slovak Republic and Zambia, out of the 90 countries and economies with available data, students’ socio-economic status accounts for 20% or more of the variation in performance (Table I.B1.2b.2). By contrast, students’ socio-economic status accounts for less than 7% of the variation in performance in 23 other countries and economies.

While a weak association between student socio-economic status and performance within countries/economies is necessary for achieving fairness in education, it is not, in itself, a sufficient condition. It is also important to consider fairness in terms of education systems’ overall levels of performance. A country/economy that combines high levels of fairness in terms of student socio-economic status with low mean performance – indicating poor achievement across the board, regardless of students’ socio-economic status – should not be viewed as a desirable outcome.

As shown in Figure I.2.12, in many cases, countries and economies with higher levels of fairness by socio-economic status are not necessarily those with strong student performance. Similar to what is observed in science, differences in students’ socio-economic status account for 9% of the variation in reading and 12% of the variation in mathematics performance on average across OECD countries (Tables I.B1.2b.3 and I.B1.2b.4).

Another way to describe socio-economic disparities is to examine the score difference associated with a given increase in socio-economic status. On average across OECD countries in 2025, a one-unit increase in the PISA index of economic, social and cultural status is associated with an increase of 36 score points in the science assessment (Table I.B1.2b.2). This is almost twice what 15-year-old students typically learn in a year (see Box I.2.2).11

The performance gap related to students’ socio-economic status is widest in the Slovak Republic, where a one-unit increase in the index is associated with a difference of 48 score points in science. In Germany, Luxembourg and Zambia, the increase in the index is associated with a difference between 45 and 47 score points. By contrast, the associated change in performance amounts to less than 20 score points in 18 countries and economies. While the slope varied between countries/economies, in all countries/economies participating in PISA 2025, more-advantaged students performed better than more disadvantaged ones, with the sole exception of Kenya (Table I.B1.2b.2).

As for science, the performance gap related to students’ socio-economic status (a one-unit increase in the index) is 32 score points for reading, and 35 score points in mathematics, on average across OECD countries (Tables I.B1.2b.3 and I.B1.2b.4).

However, socio-economic gradients do not give information about the size of performance gaps related to differences in socio-economic status between the most and least advantaged students within a country/economy. This metric is shown, instead, by the mean performance of students belonging to the top and bottom quarters of socio-economic status in a country/economy, as presented in Figure I.2.12.

On average across OECD countries, socio-economically advantaged students (those in the top quarter of the distribution in the ESCS index in their country/economy) scored 85 points more in science than disadvantaged students (those in the bottom quarter of the distribution in their country/economy) (Table I.B1.2b.2). The gap between these two groups of students is higher than 85 score points in 25 countries or economies while the gap is under 50 points in 17 countries or economies. In Bulgaria, Germany, Hungary, Luxembourg, the Netherlands*, Romania, the Slovak Republic and Zambia, the gap between socio-economically advantaged and disadvantaged students is the highest, exceeding 100 score points.

As for science, socio-economically advantaged students scored 77 points more in reading and 83 score points more in mathematics, than disadvantaged students, on average across OECD countries (Tables I.B1.2b.3and I.B1.2b.4).

A complementary way to examine socio-economic gaps is to place all PISA students on a common international socio-economic scale, rather than defining advantage and disadvantage separately within each country or economy. This shows a similar overall pattern: in all participating countries and economies with available data, students in the highest international quarter of socio-economic status perform better in science than students in the lowest international quarter (Figure I.2.13). Socio-economic disparities in science performance are therefore visible whether socio-economic status is measured within countries/economies or on a common international scale.

These average gaps provide an important summary of socio-economic disparities in performance, but they do not show how disadvantage is linked to the risk of low achievement or, conversely, the extent to which some disadvantaged students succeed despite adversity. These questions are examined next.

The strength in the relationship between science performance and socio-economic status has decreased across OECD countries (Table I.B1.2b.25). Across participating countries in PISA 2025, the strength of the relationship only increased in Indonesia, Macao (China) and the Philippines. Likewise, the strength in the relationship between both reading and mathematics performance and socio-economic status has also decreased across OECD countries (about 4 percentage points). As will be discussed next and shown in Figure I.2.14, in a number of systems, the reasons behind these decreases are possibly related to a decline in academic performance among socio-economically advantaged students while performance remained mostly unchanged among their more disadvantaged peers.

Changes in socio-economic disparities between 2022 and 2025 can be measured by the difference in average performance in science between socio-economically advantaged and disadvantaged students (hereafter, this will be referred to as the “socio-economic gap”). A narrower socio-economic gap means there is less disparity in performance between advantaged and disadvantaged students; by contrast, a wider gap indicates greater disparity.

A shrinking socio-economic gap does not necessarily mean a fairer system that cultivates and expands the opportunities for the most disadvantaged students. A smaller socio-economic gap can result from stagnating performance among disadvantaged students while the performance of advantaged students declines. This is what is observed in science, on average across OECD countries between PISA 2022 and 2025 (Table I.2.10); the socio-economic gap has been reduced by 13 score points on average, due to declining average scores among advantaged students rather than by improvements among disadvantaged students.

The performance of disadvantaged students in science remained stable between 2022 and 2025 on average across OECD countries (mean score of 447 score points) and in 47 countries/economies (Table I.B1.2b.22). Declines were observed in only 5 countries/economies (Guatemala, Hong Kong (China), Latvia, Morocco and Paraguay).

Systems that achieved greater socio-economic fairness over time are those where the performance of the most disadvantaged students improved as much, or more, than that of their most advantaged peers. As shown in Table I.2.10, disadvantaged students’ performance in science improved in 21 countries/economies; and in all of these countries, the gap with their most advantaged peers either reduced or remained stable. The increase in performance ranged from 13 points (in France, Iceland and the United Kingdom) to over 40 points (in, Georgia, the United Arab Emirates and Uzbekistan). In the latter three countries, the mean score of disadvantaged students in PISA 2022 was lower than 400 points; in other words, these disadvantaged students improved their scores starting from a performance level that was very low (Table I.B1.2b.22).

Advantaged students’ performance in science declined on average across OECD countries (by 8 score points) and in 18 countries/economies, most of them OECD countries. Advantaged students’ performance declined in all cases by over 10 score points and by more than 20 points in 5 countries/economies (Denmark, Israel, Latvia, Slovenia and Sweden) (Table I.B1.2b.22).

The performance of advantaged students in science did not change significantly in 43 countries/economies, and it improved in 12 countries (one OECD country, Türkiye, only). These are countries/economies with very different profiles, where advantaged students’ performance in PISA 2022 was either among the highest across all PISA‑participating countries/economies (Macao (China) and Türkiye), in the middle range (Georgia and Montenegro), or below the baseline (Cambodia and the Philippines); in other words, these advantaged students improved their scores starting from very different performance levels (Table I.B1.2b.22).

Given that, in many countries and economies, science performance remained stable among disadvantaged and advantaged students, it is not surprising that in most, the socio-economic gap in science performance did not change between 2022 and 2025.

Similar patterns are observed in reading and mathematics, where recent changes in the socio-economic gap also need to be interpreted in light of how performance evolved among both advantaged and disadvantaged students.

In reading, on average across OECD countries, mean performance declined significantly among socio-economically advantaged students (by 20 score points), more than among disadvantaged students (by 4 score points). As a result, the socio-economic gap shrank by 17 score points, largely because of the greater decline among advantaged students (Table I.B1.2b.23). The socio-economic gap in reading performance decreased in 37 countries and economies. In 28 of these, performance among advantaged students decreased and was likely the main driver of the reduction in the socio-economic gap. The gap increased only in the Philippines; in the remaining countries and economies, the socio-economic gap in reading did not change (Table I.B1.2b.23).

In mathematics, on average across 35 OECD countries, mean performance declined significantly among socio-economically advantaged students (by 14 score points), while the decline was not significant among disadvantaged students. As a result, the socio-economic gap shrank by 11 score points, again because of the larger decline among advantaged students (Table I.B1.2b.24). The socio-economic gap in mathematics performance decreased in 29 countries and economies. In 22 of these, performance among advantaged students decreased and was likely the main driver of the reduction in the socio-economic gap. The gap increased only in Brazil, the Philippines and Türkiye. In the remaining countries and economies, the socio-economic gap in mathematics did not change between 2022 and 2025 (Table I.B1.2b.24).

As shown in Figure I.2.15, socio-economic disadvantage is strongly associated with a higher risk of low performance in science. On average across OECD countries, 37% of socio-economically disadvantaged students, but only 12% of advantaged students, scored below proficiency Level 2 in science (a gap of 25 percentage points). The difference in the percentage of low performers in science between advantaged and disadvantaged students is 20 percentage points or more in most countries and economies; in Romania it exceeds 50 percentage points and reaches 66 percentage points in Zambia (Table I.B1.2b.14).

Nevertheless, some disadvantaged students achieve strong results. In PISA, “academically resilient students” are defined as students who are in the bottom quarter of the PISA index of economic, social and cultural status (ESCS) in their own country/economy, but who score in the top quarter of performance in that same country/economy. These students are considered academically resilient because, despite their socio-economic disadvantage, they have attained educational excellence by comparison with students in their own country.

As shown in Figure I.2.16, the percentage of academically resilient students in science varies from less than 10% in some countries/economies (Austria, Belgium, Brazil, Colombia, Ecuador, Germany, Luxembourg, Romania, the Slovak Republic, Sweden, and Zambia) to 22% or more in others (Cambodia, Kenya, Rwanda and Uzbekistan). On average across OECD countries, 12% of disadvantaged students scored in the top quarter of science performance in their own countries/economies and are considered academically resilient. In reading and mathematics, the percentage of academically resilient students is 13% and 12% respectively, on average across OECD countries (Tables I.B1.2b.3and I.B1.2b.4).

Complementing these analyses, a multidimensional deprivation indicator based on PISA 2025 data captures the extent of student deprivation in terms of a lack of essential items and activities due to financial constraints, across two distinct thresholds: material deprivation and social deprivation. Analyses show that this indicator identifies forms of hardship that, while closely related to socio-economic status, are not fully captured by it. They also show that both lower living standards and more limited opportunities for social participation are associated with weaker learning outcomes (see Box I.2.5).

Differences in performance by socio-economic status can also be examined by looking at changes in the proportions of advantaged and disadvantaged students who score below baseline proficiency Level 2 (“low performers”) and at proficiency Levels 5 or 6 (“top performers”) across PISA domains.

As shown in Table I.2.11, the percentage of disadvantaged students who scored below proficiency Level 2 in science did not change significantly on average across 35 OECD countries, although it decreased in 21 countries and economies (out of 73 with available data) between 2022 and 2025. In 9 of these countries and economies, the share of disadvantaged students scoring below Level 2 in science declined by more than 10 percentage points, and in seven it declined by more than 15 percentage points. This trend keeps the share of disadvantaged low performers at levels observed in previous cycles (2015 and 2018).

By contrast, in only three countries and economies, the share of disadvantaged students scoring below Level 2 in science increased. In all of these, over one in four disadvantaged students scored below proficiency Level 2 in science in PISA 2025, ranging from 26% in Hong Kong (China) to 33% in Latvia and 93% in Guatemala (Table I.B1.2b.28).

In 49 countries and economies, the percentage of disadvantaged students scoring below proficiency Level 2 in science did not change significantly between 2022 and 2025.

Among socio-economically advantaged students, the share of low performers in science increased on average across OECD countries (by about two percentage points) between 2022 and 2025. It rose in 12 countries and economies (mostly OECD countries), ranging from around 2 points in Singapore to more than 8 percentage points in Israel and Malta. Conversely, the share of low performers among advantaged students declined in 14 countries and economies, reaching decreases of up to 22 percentage points in Cambodia and 49 percentage points in Uzbekistan. In the large majority of countries and economies, this share did not change significantly.