B2 is closer to B1 than you think.
We tested 92 differences between the two exams and eight survived. Reading and speaking produced none at all. Here is what B2 actually asks, and where the real step up sits.
This is the Staatsexamen NT2
B2 is Programma II of the Staatsexamen NT2, set by the CvTE, and it shares its format with Programma I at B1. That matters for the comparison further down: B1 and B2 come from the same body in the same years, so a difference between them is a difference in the exam rather than in who wrote it.
Counted from the exams the CvTE publishes in its NT2 practice environment.
Reading
108 questions, 3 exams (2023 to 2025)More inference than at B1, and less that stands there literally. But not one of the reading differences we tested against B1 was large enough to publish, so treat the shift as a matter of degree.
The trap is the same one B1 leans on and it leans harder: 47 of the 108 wrong options are true statements that answer a question you were not asked. What changes is the balance. Detail questions drop below half, and inference climbs to 29 of the 108.
In the exam
- Keep asking whether an option answers this question rather than whether it is true. That trap is 47 of the 108, the largest single group.
- Expect to reason rather than locate. Only 13 of the 108 answers stand there almost literally.
- Do not skim for the gist. Detail is still the largest question type at 52 of the 108. The Staatsexamen NT2 practice exam is where the difference between skimming and reading shows up.
Listening
115 questions, 3 exams (2023 to 2025)Word recognition does not work here, and not because the answer avoids the words you heard. It uses them about as often as the wrong options do.
Two points separate the right answer from the wrong ones on word overlap, so hearing a familiar word tells you almost nothing. What decides it is whether the option answers the question that was asked, and the wrong ones are built to be true about the fragment without doing that. Two thirds of them now draw on the whole fragment rather than one passage of it, so a phrase you remember hearing is weak evidence either way.
In the exam
- Do not pick an option because you heard those words. The key echoes the fragment 85% of the time and the wrong options 83%.
- Ask whether the option answers the question, not whether it is true about the fragment. Being true is what the wrong ones are built for.
- Hold the whole fragment in mind rather than one line. Two thirds of the wrong options are assembled from across it.
- Expect monologues more often than at B1. 23 of the 115 fragments have a single speaker, against 10 of 115 one level down.
Writing
30 tasks, 3 exams (2023 to 2025)Ten tasks per exam, and two things change completely from B1. Every task now lets you invent the details, and every task has two fields to fill rather than one.
At B1 two thirds of the tasks pinned you to given facts; at B2 none of them do. That sounds like freedom and it is the opposite: inventing plausible detail that fits the situation is now part of what is being marked, and there is nothing to fall back on when you cannot think of anything.
In the exam
- Practise inventing plausible detail. All 30 tasks let you, and all 30 expect you to fill two fields rather than one.
- Work out what the situation needs before you write. In 21 of the 30 the points are not listed.
- Prepare for work and study settings above all. That is 27 of the 30 tasks, and email is 20 of them.
How writing is marked
The CvTE publishes the model. Ten tasks, 58 points, and the weighting is close to B1 with one telling change.
Scroll the table sideways to see what each criterion measures.
| Criterion | Points | What it measures |
|---|---|---|
| Adequacy | 20 | Does the text do what the task asked, for the right reader |
| Grammar | 13 | Grammatical accuracy |
| Word use | 5 | Range and choice of vocabulary |
| Spelling | 5 | Spelling of the words you used |
| Coherence | 5 | Whether the parts hang together |
Grammar weighs less here than at B1, 13 points against 16, while coherence rises from 3 to 5. The exam expects longer texts that hold together, and it is willing to forgive a slip to get them.
Adequacy is still the whole game
Twenty points of 58, more than a third, for whether the text does what the task asked of the right reader in the right tone. That is the same twenty as at B1, on an exam worth more points, so a text that misses the point loses even more here than one level down.
The pass mark is published for this exam: 31 to 34 of the 58 points, depending on the year. That is the CvTE figure and not an estimate of ours.
Source: CvTE marking models, Schrijven II, 2023 to 2025.
Speaking
39 tasks, 3 exams (2023 to 2025)Thirteen tasks, weighted towards the longer second part. Not one speaking difference with B1 survived testing, so what holds there holds here.
Two thirds of the tasks sit in the second part, where the monologues are, and that is the difference you feel even though it does not show up as one in the numbers. The instruction still tells you what is wanted: how many things to cover, and whether a reason is expected.
In the exam
- Read the instruction for what it asks. Nearly half the tasks, 19 of the 39, want a reason and leaving it out costs poinks on the heaviest criterion.
- Prepare for speaking at length. 27 of the 39 tasks sit in the longer second part.
- Be ready to answer a person. 24 of the 39 start with someone speaking to you. Drilling that is what the Staatsexamen NT2 Programma 2 course is for.
How speaking is marked
Six aspects, 132 points across thirteen tasks. Two of them cannot be judged from a transcript, which matters for how you practise.
Scroll the table sideways to see what each criterion measures.
| Aspect | Points | What it measures |
|---|---|---|
| Content | 31 | Whether you said what the task asked for |
| Word and sentence formation | 31 | Grammatical accuracy |
| Vocabulary | 27 | Range and choice of words |
| Pronunciation | 27 | Whether you can be understood |
| Tempo | 9 | Pace and hesitation |
| Word choice | 4 | Whether the words fit the situation |
Vocabulary carries more weight here than at B1, 27 points against 24, and word choice less. The exam is asking for range rather than for exactly the right word.
Pronunciation and tempo are 36 of the 132, over a quarter, and neither survives becoming a transcript. Our own feedback therefore reaches about three quarters of what a real assessor weighs, and we would rather say so than let a score here read as an exam prediction.
The pass mark is published for this exam, and the exam was rebuilt in 2024: 67 of 126 points in 2023, then 75 and 73 of 132 in 2024 and 2025. Those are CvTE figures, not ours.
Source: CvTE marking models, Spreken II, 2023 to 2025.
How B2 differs from B1
We ran 92 tests across every field we coded. Eight came out large enough to publish and survived a check on how reliably the field was coded, and all eight are in listening and writing. Reading and speaking produced none.
Listening: wrong options drawn from across the whole fragment
38 of 112 at B1 against 75 of 114 at B2
Writing: task lets you invent the details
12 of 36 at B1 against 30 of 30 at B2
Listening: fragments with a single speaker
10 of 115 at B1 against 23 of 115 at B2
The columns read B1 then B2 on this page. There is no exam-board confound here: both are the Staatsexamen NT2, from the same years, coded the same way. The difference between them is genuinely a difference in the exam, and the honest summary is that there is less of it than the level names suggest.
One thing to keep in mind
These are three consecutive years of one exam, not a sample of every paper the CvTE has ever set. Where a field was too uncertain to code, the item was left out rather than guessed, which is why some counts run to 114 rather than the full 115. We publish what we counted and what it rests on, which is also how inburgeringscursus is built.
How this was counted
Every question in three years of exams was coded on what it asks, how the wrong answers are built, and what the right one looks like. Every difference with B1 was then tested before it was published, and most of them did not hold. That is why this page argues the two levels are closer than they look, and the ones that failed are named here and on the Dutch exam research page.
Counted August 2026, covering the 2023, 2024 and 2025 exams.
Timo Yasamin · Founder, Inburgeringscursus
Timo started Inburgeringscursus after watching people prepare for this exam with material nobody had checked against the real thing. So he had it checked. Every figure on these pages is counted from the official exams, and the patterns that did not survive that check are published next to the ones that did.
Practise the B2 exam
on the real format.
One full practice exam per part, free, in the format the real one uses.