One of these certificates expires in two years

Each of these exams documents itself clearly. None of them documents itself against the others, so the questions that only appear in comparison have no answer anywhere — and two of them decide real money.

Does the certificate expire? And can a good total score still fail me? The twenty-five exams below disagree with each other on both. Every cell comes from the examining body’s own document, and every row carries the date it was read — shown in the last column, not summarised here, so a new row cannot arrive without its provenance. Take any of it and use it; no credit required.

The clocks nobody mentions

TOPIK scores are valid for two years from the date the results are announced. That is stated in the Korean Ministry of Education’s 2026 administration notice.

It is not the only clock, and this page said for several days that it was. Of the twenty-five exams here, four state a term after which the certificate stops being current: TOPIK (two years), SIELE (five years), TOEFL iBT (two years), TYS (two years). Every other body here either says the certificate never expires, declines to answer, or is silent on the point. That sentence is now rendered from the dataset rather than typed beside it, because the hand-typed version was falsified by the SIELE row on the day it landed and nobody noticed.

The rest do not work that way. The DELE has indefinite validity, per the Instituto Cervantes. The JLPT’s own FAQ is blunter still: the certificate never expires, and results from the pre-2010 version of the test remain valid.

There is a qualification, and it belongs here because it is the JLPT’s own. The same FAQ notes that companies and educational institutions referring to results sometimes set a time limit of their own. So “does this certificate expire?” and “will this university accept one from four years ago?” are different questions, and only the second decides anything. The exam body answers the first. Ask the institution the second.

The practical consequence is about sequencing rather than studying. A JLPT or DELE can be sat whenever you are ready. A TOPIK should be counted backwards from the deadline that needs it, or you will sit it twice.

Two exams, opposite failure modes

The JLPT and the TOPIK are the cleanest contrast in the table, because they resolve the same design question in opposite directions.

The JLPT applies a sectional minimum: 19 out of 60 in each scoring section at N1–N3. Its own wording is that a candidate below the sectional pass mark in even one section has failed “no matter how high the total score”. The arithmetic that follows is uncomfortable — 60 in vocabulary, 60 in reading and 18 in listening is 138 out of 180 at N2, far past the 90 required, and a fail.

The TOPIK has no sectional minimum at all. The 2026 notice states that the grade is decided by the sum of the section scores. A weak section can be carried by a strong one.

This is the rare comparison that changes what you should actually do on a Tuesday evening. Preparing for the JLPT, your weakest section is your study plan, because it alone can fail you. Preparing for the TOPIK, points are points, and the rational move is to push wherever points come cheapest. Advice imported from the wrong exam is worse than no advice.

France sits with the JLPT on design and against it on severity. The DELF and DALF apply both an aggregate pass mark and a per-skill floor — 50 out of 100 overall, and a minimum of 5 out of 25 in each épreuve, with France Éducation international stating that a mark below it is éliminatoire. That makes them and the JLPT the only two exams here that combine the two mechanisms. The floors are not comparable in strictness: the JLPT’s 19/60 is about 32%, the DELF’s 5/25 is 20%. A candidate scoring the French minimum in speaking needs 45 of the remaining 75 — 60% across the other three — and holds the diploma.

And at DALF C2 the four skills become two. Every level from A1 to C1 is marked as four épreuves of 25. FEI’s C2 page lists les 2 épreuves de l’examen: one 3½-hour paper combining written comprehension with written production, and one 30-minute individual test combining oral comprehension with oral production, each out of 50 with a 10/50 floor. The percentage floor is unchanged at 20% — what changes is the granularity. Below C2, reading and writing can fail separately; at C2 they are one mark, and comprehension is never scored on its own at all.

Italy’s CELI does the same thing at the other end. In its migrant line, the CELI Impatto i (A1) ties listening comprehension to oral production so the two costituiscono un’unica prova — one test, no separate listening paper. Listening becomes its own component only from CELI 1 i (A2) upward.

So two families here change what counts as a separately-scored skill as the ladder moves, at opposite ends and for opposite reasons. The DALF fuses at C2 because separating what a candidate understood from what they did with it is artificial at mastery level. The CELI fuses at A1 because a candidate arriving from a literacy course cannot be handed a listening paper to fill in — the paper would measure reading. The shared rule underneath is that fusion happens wherever splitting the skills would measure the wrong thing, which is a better account of what a “skill” is on a language exam than any single one of these bodies gives.

ten of twenty-five publish no price

The fee column began as a record of what is not published. That heading used to say “most of them”, typed by hand; it stopped being true on 11 September 2026, when three rows flipped in one morning because the FAQ and the booking form were read instead of the information pages — the Goethe-Institut, the SIELE and the TOEFL all publish a price after all, and the TestDaF had the day before. The count above is now derived from the table, and the sweep that found them is described under each row’s guide.

The JLPT has no global fee. It is set by the local host institution in each country. Neither does the HSK — the official test site publishes syllabus documents and test dates but no price, because fees are set by the test centre. For both, the only correct number is the one your own centre publishes, and any single figure you see quoted online is somebody’s local price presented as a global one.

TOPIK does publish, but only for Korea. The 2026 notice gives ₩40,000 for TOPIK I on paper, ₩55,000 for TOPIK II, ₩70,000 and ₩95,000 for the internet-based versions and ₩80,000 for the speaking test — roughly $29, $40, $51, $69 and $58 at European Central Bank reference rates for 25 August 2026. The same notice says overseas fees vary by country and must be confirmed locally, so those figures are Korean domestic prices and nothing more.

Portugal’s CAPLE publishes no fee at all. Its pages refer throughout to the propina and never name a figure — you register, then have 24 hours to pay by card, and the fee is non-refundable.

The Goethe-Institut publishes a full table — for Germany. This page said the opposite from 27 August to 11 September 2026. The German site’s FAQ answers Wie viel kostet ein Goethe-Zertifikat? directly and links its 2026 list: adults A1 €155, A2 €175, B1 €259, B2 €289, C1 €349, C2 €359, single modules €104 to €139, TestDaF €215, course participants 20% off. So four B1 modules cost €416 against €259 for the whole exam. The central exam pages carry no price and booking abroad runs on each institute’s own page, which we have not read: one country’s table in full, everybody else’s local.

The TOEFL and the SIELE both price by country, and both hide it behind a selector. ETS’s registration page says select your testing location; the file behind it lists 212 countries, US$170 to US$475 (Nigeria and Ukraine at the bottom, Switzerland at the top, US$255 the most common). siele.org’s booking form lists 95 countries and prices SIELE Global at 135 € to 175 €. Both were recorded here as “not published” until 11 September 2026, for the same reason: the information pages were read and the form was not.

telc says it plainly enough to end the argument. Its own FAQ states that because each examination centre sets its own examination fees, telc cannot give any information about the costs. That is an exam body saying, in writing, that it does not know what you will be charged.

Brazil’s Celpe-Bras publishes a ceiling rather than a price. Inep names R$ 247,00 for sittings in Brazil and US$ 115,00 abroad, and adds the clause that gives those numbers force — the registration fee charged by the centres may not exceed the amounts set in the edital. A centre may charge less; it may not charge more. Most other published figures in this table are a price at one centre, true of that centre and of nothing else. A ceiling bounds all of them at once, and finding it was what turned the fee column from a single failure into a set of distinct postures — counted at the end of this section, where they are enumerated rather than tallied here.

Poland takes the same posture and tiers it twice. The Państwowa Komisja publishes a maximum per level rather than a price, and lets each accredited centre set its own beneath it — wysokość opłat ustala każdy z ośrodków egzaminacyjnych. The caps are €90 for the children and youth exams at A1–B2, €120 for adult A1 and A2, €150 for B1 and B2 and €180 for C1 and C2, with a separate €20 certificate fee paid only on passing. So the same B2 paper is capped at two different numbers depending on the candidate’s age, which no other row here does. The detail is in the Polish certificate guide.

The Dutch Staatsexamen NT2 was the first one here that simply tells you the number. DUO publishes €50 per component and €200 for a complete exam, and states that this applies to Programma I and Programma II alike. Three things follow. The price is national and published rather than a range or a per-centre figure. €200 is exactly four times €50, so splitting the exam across sittings costs nothing extra and sitting it all at once earns no discount. And the B2 exam costs the same as the B1 one — the fee attaches to the act of examining rather than to the level examined, which is the opposite of the DELE, where the higher levels were dearer at every centre we read. DUO does warn that paying from abroad often costs more.

Finland’s YKI publishes a national price too, and disagrees with the NT2 about what a price should attach to. OPH lists €165 basic, €190 intermediate, €216 advanced, at 0% VAT. So where DUO charges the same for B2 as for B1 because the fee attaches to the act of examining, the YKI charges €51 more at the top because the fee attaches to the level. Two exams that both answer the question plainly, and answer it differently.

The YKI is also the only exam here that prices the four skills differently from each other. The NT2 prices its four components too, but at a flat €50 apiece. For a Finnish or Swedish intermediate test taken as individual subtests, OPH publishes listening €43, reading €43, writing €70 and speaking €84. Two things follow, and both are unique in this table.

First, the productive skills cost about twice the receptive ones. Assessing someone’s speaking costs the examining body roughly double what assessing their reading costs, and the fee table is the only place in this dataset where that distinction is priced instead of asserted.

Second, those four subtests sum to €240 against €190 for the complete test — buying the parts costs €50 more than buying the whole. That is the exact inverse of the NT2, where four components at €50 come to precisely the €200 of a complete exam. The NT2’s modularity is price-neutral; the YKI’s is priced against you.

So the fee column now records twelve distinct postures, not one failure: no published figure (JLPT, HSK, CAPLE, DELF / DALF, IELTS, Swedex), published per centre (DELE, ТРКИ / TORFL), a full price table for the home country; abroad, each institute's own page (Goethe-Zertifikat), an explicit disclaimer (telc, Cambridge English), a flat national price that ignores level (Staatsexamen NT2), a legal maximum (Celpe-Bras, Certyfikat (state)), a national price that rises with level (YKI), tiered by country group, only the home tier published (TestDaF), one central price list, tiered by the country you sit in (SIELE, TOEFL iBT, TYS), devolved entirely to local government (Norskprøven, Norskprøven C1), free for nationals, indexed to a welfare benchmark for foreigners (Держсертифікат (state)), and one published price at home and one abroad (Tisus).

France Éducation international is the most helpful of the silent ones, and still unquotable. Its practical-information page answers the whole fee question in a sentence — le centre d’examen me renseigne sur les coûts d’inscription — so it does not merely omit the price, it names the party holding it. A candidate still cannot know the number before contacting somebody.

So ten of the twenty-five exams here publish nothing globally — JLPT, HSK, telc, CAPLE, DELF / DALF, Norskprøven, Norskprøven C1, IELTS, Cambridge English, Swedex — which means any single number circulating online for one of them is somebody’s local price, at some past sitting, presented as a global one.

The DELE publishes per centre, and the spread is the largest thing in this table. Reading four Instituto Cervantes centres’ own 2026 fee pages, London was dearest and New Delhi cheapest at every one of the six levels — a spread of roughly three to three-and-a-half times for the identical paper. The detail, the caveats and why it is not arbitrage advice are in the DELE guide.

Nine levels, six levels, and levels you cannot choose

The level columns are not comparable and should not be read as if they were. A “level 4” in one exam has no relationship to a “level 4” in another, and even within a single exam the scales can differ: TOPIK is marked out of 300 on paper and 600 online, with different cutoffs for the same grade.

Two structural facts are worth carrying anyway. On the TOPIK you do not register for a level — you register for a test, and TOPIK II awards levels 3 to 6 only, so 119 out of 300 returns no grade at all where TOPIK I would have returned a level 2.

The Celpe-Bras goes further: there is no test to choose either. One exam, sat by everybody, with the level assigned from performance afterwards — so it is the one exam here that cannot be failed at a level, only certified lower or not at all. Its band is then set by the lower of the written and oral scores rather than a total or an average, which makes it the strictest sectional rule in the table: a top written mark is worth nothing above what the weaker half supports. That is a fourth position on sectional scoring, distinct from the JLPT’s floor, the TOPIK’s absence of one, and the Goethe modules that never combine. It also has no certificate below intermediate at any score — a distinction it now shares with exactly one other exam here, which is otherwise its opposite in every respect.

The Dutch Staatsexamen NT2 has only two levels in the entire system: Programma I at B1 and Programma II at B2. Nothing certifies elementary Dutch and nothing certifies mastery. Where the Celpe-Bras has no floor but runs to Avançado superior and assigns your band from the weaker of two halves, the NT2 has no floor and no ceiling above B2, you choose your programme deliberately, and its four parts never interact — no aggregate, no compensation, each passed or failed alone.

And it answers the reassembly question the Goethe raises, in the opposite direction. Both split into four separately-sat components that each yield their own certificate. But Goethe requires all four within one year, and within the EU, one country, or one cooperation partner centre, for a single combined certificate — splitting is free, reassembling is not. DUO states that de certificaten en het diploma zijn onbeperkt geldig: the component certificates and the diploma are valid without limit. No window, no geography, no clock running against your earlier passes. On our first-party reading that is the most permissive modular structure in this table, and the comparison is ours rather than either body’s.

And the HSK now has nine levels, under a syllabus that took effect in July 2026, taken as seven exams. Counting its own word lists gives 11,000 entries, of which more than half sit above HSK 6: the top band alone adds 5,600 new words against the 5,400 in all of levels 1 to 6 combined. Guides still describing six levels are describing the standard it replaced.

German is the one language here with two exam families in the table, and they disagree about the most basic question of all: how much of the exam you may sit at once. From B1 upward the Goethe-Zertifikat is four separate modules, sat individually, at different centres, each producing its own certificate. telc allows the written and oral parts on different days and no more than 7 days apart, with no module system at all. Same CEFR level, same recognition, and opposite answers — set out in the telc-versus-Goethe comparison at B1.

A wrinkle worth carrying: the ÖSD-Zertifikat B1 is not a third option. The Goethe-Institut states on its own page that the B1 exam was developed jointly with the ÖSD and the University of Freiburg in Switzerland, and is offered worldwide under either name. One exam, two certificates.

Which exams test speaking at every level

This is the split that decides what a candidate should actually practise, and it runs the opposite way to most people’s assumption.

Portugal’s CAPLE certificates score speaking at every level, A1 to C2 — a component of the paper you register for, not an add-on. At A1 it is video-recorded and assessed both by the exam centre and by CAPLE in Lisbon.

So do the Goethe-Zertifikate, at every level from A1 to C2 — speaking is one of the four modules, and above B1 it is a separately certificated examination in its own right. The format changes as you climb, which the durations hide: A1 is a group oral, A2 a pair, B1 to C1 are sat with a second candidate, and C2 is an individual examination. So the fifteen minutes listed at B2 are shared and the fifteen listed at C2 are not.

The JLPT has no speaking section at any level. Neither does TOPIK I or TOPIK II — Korean speaking is an entirely separate exam, with its own fee and only three sittings a year against six for the written papers. For the HSK, the official body lists the spoken test (HSKK) separately from the HSK itself.

So a Japanese or Korean candidate can hold a strong certificate that says nothing directly about whether they can talk to anyone, and a Portuguese or German candidate cannot. This is the single most useful thing to know before choosing what to buy, and we have said it in the direction that costs us on the JLPT page and in the direction that suits us here — check both against the exam bodies’ own component lists rather than against us.

Why only first-party pages — an exam body’s own words

Every figure on this page was read from the examining body’s own document. That rule costs us: it is why French is blank, and why several cells say not verified rather than carrying a number we could easily have copied from a test-prep site.

The best defence of it we have found is written by an exam body about itself. Italy’s CILS centre, at the Università per Stranieri di Siena, puts this on its own pages: the CILS Centre is responsible for the information on these pages, and news or communications provided by other sites are not managed by the CILS Centre and may therefore be unreliable.

An examining body telling you, unprompted, that second-hand information about its exam may be wrong is about as clear a statement of the problem as exists. It is also the reason we would rather publish a gap than a plausible number.

CILS itself is reachable and not yet written up here — it is the first Italian certification to adopt the six-level CEFR scale, an institutional member of EALTA and an affiliate of ALTE. Its per-level pages return a server error to us at the time of writing, so rather than build a row out of the parts that did load, it is named here as pending. That is the same call the gaps below get.

What is missing, and why

Two exam families used to be absent from this table. Both are now rows. The rule that kept them out has not changed: a number goes on this site only if we read it on the examining body’s own page.

DELF and DALF were the first gap, and closed on 2026-08-30. This page previously said that France Éducation international serves an automated-access challenge to our reader, that working around it is precisely what it exists to prevent, and that we therefore had no first-party figure for the French diplomas. The first half is still true of our fetcher. What changed is that we stopped treating a refusal to one client as a property of the source: FEI’s pages answer an ordinary browser normally, and the levels, durations, pass marks, floors and fee posture in the French row were read there on 30 August 2026. No wall was worked around — a different door was already open. The gap was ours, not theirs, and it had stood for four days.

The German exams were the second gap, and they are now a row. This page previously said that goethe.de refused our requests and that every Goethe figure available to us was second-hand. That was true, and it was fixed by a teacher rather than by us. a German teacher answered a cold email asking what first-party route we were missing, and pointed us at the Goethe-Institut USA exam portal, which does answer. She was credited here by name and has asked for the credit to be taken down, so it has been. She has no affiliation with KoloLingo. The German row and the guide built from it were read on 27 August 2026.

One thing did not change: the fee cell still says not published. The Goethe-Institut attaches fees to date-specific booking listings rather than a general table, so there is no global figure to print — the gap narrowed to exactly its real size rather than closing entirely. This is what we mean by asking to be corrected: the offer is open on every page here, and it has now been taken up once.

Cells marked not verified follow the same rule in miniature. We did not find a validity period or a sectional-minimum rule stated in the HSK syllabus, and we have not confirmed the DELE’s sectional requirement first-party. Rather than infer them from how similar exams behave, they are left empty. If you know where those are published, tell us and we will fill them in and say who did it.

Where we stand in this

We run a marketplace for language teachers, so treat this as interested rather than neutral. The defence is that none of these figures are ours — every one is checkable on someone else’s site — and that the findings which cost us are on the pages rather than left off them.

The clearest of those: for the JLPT and for HSK levels specifically, conversation lessons are not the efficient purchase. Neither exam tests speaking, and we sell conversation. Where a teacher genuinely earns their fee is narrower and more honest — the written production on TOPIK II and the DELE’s writing and speaking papers, because those are the parts you cannot mark yourself.

And nobody can promise you a pass on any of these — not a school, not an app, not a teacher, not us. Corrections to hello@kololingo.com and we will publish who corrected us.

Sources

JLPT — pass marks, sectional minimums and certificate validity from the JLPT’s own FAQ and its published scoring guidelines. TOPIK — level cutoffs, fees, the grade rule and the two-year validity from the Korean Ministry of Education’s 2026 administration notice (공고 제2025-317호), hosted on topik.go.kr. DELE — levels, indefinite validity and per-centre fees from the Instituto Cervantes and four of its centres’ own fee pages. HSK — level structure and word lists counted from the official exam syllabus published by the Center for Language Education and Cooperation, linked from chinesetest.cn. CAPLE — the certificate ladder, its CEFR mapping, the component structure, the 16-year age floor, the 24-hour non-refundable fee rule and the 2020 exam statistics from CAPLE at the University of Lisbon.

Currency conversions use European Central Bank reference rates for 25 August 2026.

The other datasets on this site: what lessons cost and what teachers keep.