PISA 2025 Results: Global Overview & Key Findings

Reading showed the clearest short-term deterioration. Across 35 OECD countries with comparable data, the average reading score fell by about 14 points from 2022 to 2025, while mathematics fell by 9 points. Science did not show a statistically detectable OECD-wide decline over the same three-year period.

The longer view is more severe for reading and mathematics. From 2015 to 2025, the OECD average fell from 489 to 461 in reading and from 485 to 463 in mathematics. Science moved from 489 to 482 over the same decade.

B-S-J-Z (China) and Singapore were the two strongest systems across the three traditional PISA domains as a group, but the domain leaders differed. B-S-J-Z (China) recorded 597 in science and 612 in mathematics, while Singapore led reading at 535. Macao (China) led the new computational problem-solving scale at 572.

Eight systems were among the top 10 by reported mean score in science, reading and mathematics: B-S-J-Z (China), Singapore, Estonia, Japan, Korea, Macao (China), Chinese Taipei and the United Kingdom. This cross-domain pattern provides a broader view than treating one subject’s score order as a complete description of performance.

The largest reported science increase in this group was Uzbekistan at 83 points, followed by Georgia at 38, Cambodia at 35, Jordan and Montenegro at 32 each, and the United Arab Emirates at 26. Among OECD countries, Costa Rica, the Slovak Republic, Türkiye and the United Kingdom were specifically identified as having higher science scores than in 2022.

Nearly one in five students across OECD countries performed below Level 2 in science, reading and mathematics simultaneously, compared with about one in six in 2022. This cross-domain measure is more demanding than being below Level 2 in a single subject: it identifies students who did not reach the PISA baseline in any of the three core domains.

The decline was steeper among lower-achieving students

The distribution widened as well as shifted downward. In the OECD trend series, students at the 25th percentile lost more ground from 2015 to 2025 than students at the 75th percentile in all three core subjects.

Science also showed a wide within-country performance spread. The OECD-average gap between the 90th and 10th percentiles was about 258 score points in 2025. Average score alone therefore does not describe how evenly performance is distributed within a system.

The OECD reading average fell by 14 points from 2022 to 2025 and by 28 points from 2015 to 2025. The deterioration was not limited to one type of reading task. OECD analysis found declines in skills that require students to evaluate information, connect information across sources and interpret texts critically.

Response behavior also changed. The OECD overview reports that the share of “hasty readers” — students who move through text rapidly and answer inaccurately — nearly doubled between 2018 and 2025 to about 9%. More detailed item analysis also found larger difficulties with longer texts and later sections of the assessment, patterns consistent with weaker persistence or attention during demanding reading tasks.

Macao (China) scored 572, Singapore 563, B-S-J-Z (China) 560 and Japan 557. Computational problem solving was strongly related to the traditional subjects but was not the same construct: the OECD-average within-country correlation was 0.76 with science, 0.75 with mathematics and 0.69 with reading.

Some systems performed markedly better in computational problem solving than their science results alone would predict. Hong Kong (China), Malaysia, Brunei Darussalam, the Philippines, Macao (China) and Iceland were all more than 30 points above the performance expected from their science scores in OECD’s analysis.

Students who reported not using AI for specific schoolwork tasks often scored higher than students who used it for those tasks. That pattern cannot be interpreted as proof that AI use lowers performance. Student background, the reason for using AI, the difficulty of the work and the way AI is used can all differ between users and non-users.

A more specific pattern is more informative: among frequent AI users, students who also reported learning how to evaluate the quality of AI-generated information tended to perform better than frequent users who did not receive that type of guidance. OECD also found that clear school rules on mobile-phone or digital-device use were associated with less reported distraction, while the direct relationship between phone bans and PISA performance remained unclear.

Environmental concern was widespread, but students were less confident about how to turn that concern into coordinated action. The OECD also found that self-reported environmental awareness declined from 2015 to 2025 across all five topics that were measured in both cycles: nuclear waste fell by 9 percentage points, greenhouse gases and deforestation by 6 points each, and water shortages and biodiversity loss by 4 points each.

Across OECD countries, student socio-economic status accounted for about 12% of within-country variation in science performance. The 2022-2025 narrowing of some socio-economic score gaps did not necessarily reflect faster gains among disadvantaged students; OECD reports that in many cases it was driven largely by declines among advantaged students.

Several systems combined above-OECD-average science performance with relatively high inclusion and socio-economic fairness. OECD highlighted B-S-J-Z (China), Canada, Estonia, Ireland, Japan, Korea and Macao (China) in this group. These comparisons show why a single mean score cannot fully describe the distribution of opportunity within a system.

The September 2026 release already reports computational problem-solving performance, but the full analysis of students’ self-regulated learning processes in digital environments is scheduled for 2027. The optional foreign-language results are also separate from the core science, reading and mathematics results presented here.