PISA 2025: Estonia Held. The Rest of the North Did Not.

The OECD published the ninth round of its Programme for International Student Assessment on Tuesday morning. Ninety-one countries and economies. More than 760,000 students tested, standing in for roughly 33 million fifteen-year-olds. The largest PISA ever run.

It is also the worst.

PISA 2025 recorded the lowest OECD averages ever observed in all three core domains — science at 482, mathematics at 463, reading at 461. One in five fifteen-year-olds is now a low performer in science, mathematics and reading at once, up from 16% three years ago.

For a region that has spent a quarter of a century being held up as the answer, it makes for difficult reading. Only one of the eight Nordic and Baltic countries came out of this cycle in good shape, and it is not a Nordic one.

In brief

• Estonia is 6th in the world in science and first in Europe in both science and mathematics — but posted its lowest reading score ever recorded.
• Latvia lost 45 points in reading in a single cycle, the steepest fall in the region.
• Denmark's education minister called it a learning crisis and convened a crisis meeting the same day.
• Norway recorded its worst reading result ever — and carries a formal notation from the OECD for excluding 10.4% of its eligible students.
• Lithuania is the region's quiet success, above the OECD average in all four assessed domains for the first time.
• Six of the eight scored at or above the OECD average in the new digital domain, including Latvia — which is 31 points below average in reading.

The scoreboard

Science was the major domain this cycle, so start there. Scores with the change against PISA 2022 in brackets:

1. Estonia — 527 (+1, not significant)
2. Finland — 504 (−7, not significant)
3. Lithuania — 488 (+3, not significant)
4. Sweden — 485 (−8, not significant)
5. Denmark — 478 (−16)
6. Norway — 470 (−8, not significant)
7. Latvia — 468 (−25)
8. Iceland — 441 (−6, not significant)

OECD average: 482.

In mathematics: Estonia 508 (−2), Denmark 471 (−19), Lithuania 470 (−5), Finland 469 (−15), Sweden 464 (−18), Latvia 460 (−23), Norway 452 (−17), Iceland 450 (−9). OECD average 463.

In reading: Estonia 499 (−12), Finland 474 (−16), Sweden 466 (−21), Lithuania 464 (−8), Denmark 460 (−29), Norway 453 (−24), Latvia 430 (−45), Iceland 422 (−14). OECD average 461.

Two patterns fall out immediately.

The first is that science held and reading did not — and that holds almost everywhere, not just here. The second is that the ordering is not the one most people carry around in their heads. Estonia leads Europe. Finland is a distant second in this region. Four of the eight are now below the OECD average in reading.

One caution before going further, and the OECD insists on it: many of these countries are statistically tied. Sweden's 485 in science is indistinguishable from the Netherlands' 485. Lithuania's own national agency published its position as a range — 20th to 29th of 91 — which is the honest way to do it. Treat single ranks as approximations, and a three-point gap as no gap at all.

One country held

Estonia is 6th in the world in science and 2nd among OECD members, behind Japan. It is first in Europe in science and in mathematics, and second in Europe in reading, one point behind Ireland.

It is the only system outside East Asia in the global top six.

The depth is what makes it unusual. Roughly 90% of Estonian students reach at least baseline proficiency in science, against an OECD average of 74%, while 11.3% are top performers against 7.2%. Only 9.9% are low performers — comfortably the best figure in this region. Estonia is not carrying a strong average on the back of an elite tail. Almost everybody clears the bar.

Then the crack.

Estonia's reading score of 499 is the lowest it has ever recorded in PISA — down 12 points on 2022 and 24 on 2018, which was Estonia's own peak year. It is the second consecutive cycle in which Estonian reading has fallen.

Education Minister Kristina Kallas led on both halves. On the good news: “the strength of the Estonian education system is that good knowledge is not the privilege of only a small group of young people.”

On the bad: “the decline in reading skills is a problem to which we must pay more attention than before. In the age of artificial intelligence, the ability to understand a text, evaluate information, connect different sources and form one's own thought becomes even more important.”

Gunda Tire, who runs international studies at Estonia's Education and Youth Board, was blunter: “This is undoubtedly a festive day, because compared with others we are in a very good place, but at the same time it also sets off an alarm bell.”

One Estonian divide deserves naming, because it rarely travels abroad. Students taught in Russian score 47 points lower in science than their Estonian-language peers — more than two years of learning. Estonia's celebrated equity is real across family income. It is much less real across language of instruction.

One country is still living on a reputation

Finland remains 12th in the world in science at 504, well above the OECD average. The education ministry's framing — that only nine systems scored significantly higher — is fair enough.

But mathematics fell 15 points this cycle and reading 16, and the long arc is the real story.

Finnish reading has gone 546 in 2000 → 547 in 2006 → 524 in 2012 → 520 in 2018 → 490 in 2022 → 474 in 2025. That is a fall of roughly 73 points from the peak. PISA's own rule of thumb puts twenty points at about a year of schooling. Mathematics has gone from a peak of 548 in 2006 to 469, a comparable collapse.

Two findings from the Finnish national team matter more than the scores.

Arto K. Ahonen, who directs Finland's PISA research at the University of Jyväskylä, reports that only 41% of Finnish students say they try harder when a task becomes difficult — the lowest share of any comparison country. That is not a reading problem or a maths problem. It is a disposition problem, and it will show up in every domain.

The second finding is stranger. In Finland, the decline is now steepest among children from advantaged families — reversing the pattern of every previous cycle.

That is not a Finnish quirk. The OECD found the same thing internationally and said so plainly: “Between 2022 and 2025, the narrowing of socio-economic gaps largely reflected declines among advantaged students rather than improvements among disadvantaged students.”

Equity gaps are closing. They are closing because the top is coming down.

Education Minister Anders Adlercreutz: “Although learning outcomes have declined and performance gaps between students have widened, Finland continues to perform well by international standards.”

One country quietly got it right

Lithuania is the only country in the group that did not really fall. Science up 3, mathematics down 5, reading down 8 — and, for the first time, above the OECD average in all four assessed domains.

Rita Dukynaitė, Lithuania's member of the PISA Governing Board, called the results “encouraging.” Vice-Minister Jonas Petkevičius added the obvious next task: “We need to improve students' reading skills.”

Lithuania's socio-economic gap also narrowed sharply, from 88 points in 2022 to 63. The OECD's caveat about why gaps narrow applies here as everywhere — but Lithuania's absolute scores held, which is what separates it from the countries below.

Its real inequality is geographic: a 65-point science gap between the capital and large cities on one side and small towns and rural areas on the other. The OECD average for that gap is 33.

And one fell off a cliff

Latvia's numbers are the hardest in this dataset. Reading fell 45 points in a single cycle. Science fell 25, mathematics 23. Latvia now sits 37th of 91 and below the OECD average in every core domain.

41% of Latvian fifteen-year-olds do not reach the minimum level in reading. Just 2.6% reach the top level, against an OECD average of 5.8%.

Education Minister Ilze Indriksone called the results shocking.

But the most useful thing said in Latvia on Tuesday came from Rita Kiseļova, who leads the PISA project at the University of Latvia, and it belongs in any responsible account of these numbers.

This cohort, she pointed out, “experienced three major reforms in nine years: curriculum changes, transition to Latvian-only education, and exam system reforms, plus Covid-19 remote learning during grades 4–6 when foundational skills develop.”

Whatever one thinks of any of those reforms individually, running all three through a single cohort — a cohort that also lost two years of primary schooling to a pandemic — is an enormous intervention. The result is now measurable. Kiseļova argued for system-level solutions rather than expecting teachers with thirty students each to absorb the problem.

There is one more Latvian number worth sitting with. On paper, Latvia is among the most equitable systems in the region: socio-economic background explains just 5.5% of the variation in outcomes, against an OECD average of 11.6%.

That is not equity. That is levelling down. Estonia is the only country in this region that combines low inequality with high performance.

The Nordic four

Denmark, Norway, Sweden and Iceland all fell, and each fell in a slightly different way.

Denmark: a learning crisis, officially

Denmark had the sharpest single-cycle fall among the Nordics — reading down 29, mathematics down 19, science down 16. It is now below the OECD average in science, which would have been unthinkable a decade ago, and these are its lowest results since it first participated in 2000.

Measured from 2018: reading down 41 points, mathematics down 40. Nearly a third of Danish fifteen-year-olds are now low-performing readers, double the 2018 share.

The political reaction was the strongest of any country. Children and Education Minister Magnus Heunicke said PISA “establishes that we are in the middle of a very serious learning crisis… It is an enormous societal problem.” He called a crisis meeting the same day. The prime minister held a press conference.

Norway: the worst reading result it has ever recorded — and an asterisk

Norwegian reading fell 24 points to 453, the country's lowest ever. Mathematics fell 17.

The proficiency picture is worse than the averages suggest. Since 2015 the share of Norwegian students below basic level has gone from 15% to 34% in reading and 17% to 38% in mathematics. Among Norwegian boys, 41% are low performers in reading.

Education Minister Kari Nessa Nordtun: “The results are very serious and confirm a development we have seen over a long period.” Morten Rosenkvist, who heads the Directorate for Education and Training: “I think the most serious thing is that more and more students are performing at the lowest levels.”

And then there is the asterisk, which Norwegian journalists led on and almost nobody else picked up.

Norway is one of six countries carrying a formal notation in the OECD's tables for sampling — alongside Canada, the Netherlands, New Zealand, the United States and Albania. 10.4% of eligible Norwegian tenth-graders were excluded from the assessment. The OECD's maximum standard is 5%, and Norway has exceeded it continuously since 2009: 2.7% in 2000, 5.9% in 2009, 6.7% in 2015, 7.3% in 2022, 10.4% now.

Fredrik Jensen, Norway's PISA project leader, drew the conclusion that matters: had participation been better, the results would likely have been weaker still. On his estimate, holding the 2015 exclusion rate would have cost roughly four points.

Sweden: record lows, and a widening class gap

Swedish reading fell 21 points to 466 and mathematics 18 to 464, both record lows. Science was stable. 31% of Swedish students are now below basic level in reading (24% in 2022) and 34% in mathematics (27%).

Sweden is the region's equity outlier, in the wrong direction. Socio-economic background explains 12% of the variation in Swedish outcomes and the gap between the top and bottom quarters is 92 points — both worse than the OECD average, and both far worse than any of the other seven countries here.

Anna Castberg, Skolverket's head of analysis, gave the figure that makes it concrete: 49% of Swedish students from less advantaged backgrounds fail to reach basic level in mathematics, against 19% of advantaged students.

The Swedish government read the results as vindication. Education Minister Simona Mohamsson: “The results show how important the government's direction for school policy is. We are now ensuring that schools focus on the basics, with more books and fewer screens in classrooms.” Minister for Higher Education Lotta Edholm added that they “confirm that the government's focus on taking schools from screen to binder is absolutely necessary.”

That claim needs a date attached. We will come back to it.

Iceland: the steepest decline of any participating country

Iceland now has the weakest results in this group by some distance: reading 422, science 441, mathematics 450.

38.4% of Icelandic students are low performers in science against an OECD average of 25.7%, and just 2.7% are top performers against 7.2%. Only 56% reach basic proficiency in reading, down from 78% in 2015. Among pupils of foreign origin, 28% do.

On the OECD's decennial measure, Iceland's average rate of change in science — −37 points — is the largest decrease of any participating country in the world.

The broadcaster RÚV called the results the red card in academic achievement, and reported that Icelandic girls declined significantly in all three domains since 2022, reversing the country's mathematics gender gap. Education and Children's Minister Inga Sæland called the situation unacceptable and announced a focus on core subjects, a reduction in the number of subject areas, a curriculum review and phone-free measures.

The result nobody expected

PISA 2025's innovative domain was Learning in the Digital World, reported in this first volume as a scale called computational problem solving, with the OECD average set at 500. The detailed analysis is not out yet.

The headline scores are the most counter-intuitive thing in the entire release. Six of the eight Nordic and Baltic countries are at or above the OECD average:

• Estonia 541 — 2nd among OECD members, 7th globally
• Lithuania 512
• Finland 507
• Latvia 507
• Denmark 506
• Sweden 503
• Norway 491
• Iceland 490

Read the Latvian line again.

Latvia scored 430 in reading — 31 points below the OECD average, among the weakest results in Europe. And 507 in computational problem solving, seven points above it. The OECD quantifies the mismatch: 66.1% of Latvian students performed better in computational problem solving than their science scores would predict, against an OECD average of 56.3%. Denmark shows the same shape: below average in science, above average in the digital domain.

There is a genuine finding buried in that, and it is uncomfortable for everybody's talking points. These are students who read poorly and reason computationally well.

Screens, phones, and what this release does not prove

The OECD was direct about screens. Secretary-General Mathias Cormann: “In reading, more screen time and less reading for pleasure have gone hand in hand with weaker results. We're also seeing more hasty reading, students rushing through a text and giving a quick wrong answer.”

The share of those hasty readers roughly doubled, from about 7% in 2018 to 11% in 2025. Andreas Schleicher, who runs the programme, put the consequence sharply: “Success has declined fastest on reading tasks particularly important in this AI age.”

Digital distraction was associated with lower science scores in 59 of the 82 systems examined. More than one in four OECD students report classmates being distracted by devices in most or every science lesson — 36% in Denmark. Students who regularly use AI chatbots for summarising, drafting and research score about 20 points lower in science, roughly a year of learning; in Estonia, 58% use chatbots for schoolwork at least weekly, against 46% OECD-wide.

The OECD does not claim that last relationship is causal, and neither should anyone else.

Two things follow that the political reaction has mostly skipped.

First, the OECD stopped short of endorsing smartphone bans in schools. It did not hand that argument a verdict.

Second, and more importantly: PISA 2025 was administered in spring 2025. Sweden's nationwide phone ban was legislated to begin in 2026. Denmark's moved through in 2025. Norway's and Finland's restrictions came in across 2024 and 2025.

The data therefore largely predate the bans, and cannot be used to evaluate them in either direction. A minister who says these results prove the phone policy was right is making a political argument, not reporting a finding.

Three caveats worth carrying

The science framework changed. PISA 2025 substantially revised its science framework, shifting focus from scientific literacy to the general outcomes of science education and adding a competency on using scientific information for decision-making. Reading and mathematics frameworks were unchanged from 2022. That cuts directly against the one comforting finding in the Nordic results: “science was stable” rests on a comparison carrying more methodological weight this cycle than the reading and mathematics declines do.

The test adapted to the student. PISA 2025 used multi-stage adaptive testing, in which the blocks of questions a student sees depend on how they performed on earlier blocks. Not every student answered the same questions.

Ranks are ranges. The OECD publishes, for each country, the list of countries whose scores are not statistically different from its own, and directs readers to a formal range-of-ranks annex.

What this means from here

The Nordic education model has been an export product for twenty-five years — the thing delegations flew to Helsinki to study.

On these numbers, four of the five Nordic countries are now at or below the OECD average in reading. The one that still isn't, Finland, has lost roughly three and a half years' worth of reading attainment since 2000.

Estonia is now the system that works, and it is worth being precise about why: near-universal baseline proficiency rather than an elite tail, a low socio-economic gradient, and enough top performers to sit sixth in the world. It is also, on Tuesday's evidence, not immune. Its reading score has fallen in each of the last two cycles.

The rest of the release is still to come. Volume II, with the detailed Learning in the Digital World analysis, has been reported as due in December, though we have not been able to confirm that date with the OECD directly. Finland's optional English-language assessment reports in spring 2027.

Sources: OECD, PISA 2025 Results (Volume I): Future-Ready Students, published 8 September 2026; national releases and reporting from Estonia's Ministry of Education and Research and ERR, Finland's Ministry of Education and Culture and the University of Jyväskylä, Sweden's Skolverket and regeringen.se, Norway's Udir and the University of Oslo, Denmark's Ministry of Children and Education, Iceland's RÚV, Latvia's LSM and Lithuania's National Agency for Education.