Cambridge Speaking Criteria by Level: B1, B2, C1, C2

Yesterday we published a finding drawn from nine Cambridge candidates: pronunciation was the highest or joint highest mark for every single one of them.

We then read every published commentary at two more levels. The pattern does not survive. Across all 23 published candidates, pronunciation is the single highest mark just three times. At B1 Preliminary it is the single highest for nobody at all.

SpeakShark is free to practise the criterion that actually separates candidates at your level, three AI conversation sessions a day with no card.

In this guide: what breaks · all 23 scores · the clearest case · B1 interaction · C2 pronunciation · why it shifts · what to practise · limits · method · FAQ

Key takeaways

Across 23 published candidates from B1 to C2, pronunciation is the single highest mark three times, and sits at the bottom of the profile four times.
At B1 Preliminary, pronunciation is the single highest mark for none of the nine published candidates. Interactive Communication is the single highest for four and top or joint top for eight.
Cambridge publishes a pair who sat the same test where the candidate with weaker grammar and pronunciation scores higher on interaction, and they finish level on Global Achievement.
At C2 Proficiency, pronunciation is at the joint bottom of the profile for two candidates out of five.
The C2 candidate marked 2.5 for pronunciation is still described as "almost always intelligible". The mark is about variation, not clarity.

What the last guide claimed and what breaks it

Our previous guide on Cambridge speaking scores read the examiner commentaries for B2 First and C1 Advanced. Nine candidates, and pronunciation sat at or above every other criterion for all nine.

That guide closed with an explicit limit. It said we had not read B1 Preliminary or C2 Proficiency, and that nothing in it should be assumed to carry to those levels.

We have now read both, in full. The limit turned out to be the interesting part.

The nine candidate finding stands exactly as published, for B2 and C1. What does not stand is the version learners will repeat, which is the levelless one: "pronunciation is never your problem in a Cambridge speaking test". At two of the four levels, that sentence is wrong.

Every published score from B1 to C2

Every number below comes from a Cambridge examiner commentary PDF. Bold marks each candidate's highest score.

Level	Candidate	Grammar	Vocabulary	Discourse	Pronunciation	Interaction	Global
B1	Cristina	4.5	4.5	4.5	4.5	5	4.5
B1	Andrea	4.5	4.5	4.5	4.5	5	4.5
B1	Laura	3	3	3.5	3	3.5	3
B1	Claudia	3.5	3.5	3.5	3.5	4	3.5
B1	Yaheli	4.5	4.5	4.5	4.5	4.5	4.5
B1	Fernanda	4.5	4.5	4.5	4.5	5	4.5
B1	Lucia	4.5	4.5	5	5	5	5
B1	Kenza	4.5	4.5	4.5	4.5	4	4.5
B1	Mohammed	4	4	4	4	4.5	4.5
B2	Enzo	4	4	4	4.5	4	4
B2	Natalia	3.5	3.5	3.5	4	4	4
C1	Gio	5	5	5	5	5	5
C1	Pedro	4.5	5	5	5	5	5
C1	Isabela	4	4.5	4.5	5	5	5
C1	Antonio	4.5	4.5	4	4.5	4.5	4.5
C1	Marta	4	3.5	4	4.5	4	4
C1	Teresa	3.5	4	4	4	4	4
C1	Anna	2	2	2	2.5	2.5	2
C2	Manuel	5	5	4	5	4.5	5
C2	Jules	3.5	3.5	3.5	4	3.5	3.5
C2	Thiago	5	5	5	5	5	5
C2	Laura	2.5	2.5	3	2.5	3	3
C2	Alejandro	2.5	2.5	2.5	2.5	2.5	3

At B1 and B2 the two grammar columns hold the same figure, because those levels score them as one criterion, Grammar and Vocabulary. C1 and C2 report them separately, which is why those rows can differ.

Read the pronunciation column top to bottom. In the nine B2 and C1 rows it is bold nine times out of nine. In the nine B1 rows it is never the sole highest. In the five C2 rows it is twice the lowest number in the row.

Kenza and Mohammed sat the same test

The cleanest evidence in the whole set is a single pair, because everything else is held constant. Same test, same partner, same examiners, same day.

Kenza beats Mohammed by half a band on grammar and vocabulary, on discourse management and on pronunciation. The commentary backs the numbers up. Her language is "very accurate" and "she makes very few mistakes". Her pronunciation is "very clear".

Mohammed is the weaker speaker on paper. The examiner notes "some instances of unclear language" and inaccuracies in what he says.

Mohammed still scores half a band above Kenza on Interactive Communication, and the commentary is blunt about why. He "contributes well to the discussion, interacting effectively with his partner and often moving the discussion forward". The quoted example is not sophisticated English. It is "What about this bag, what do you think about this bag?"

Kenza's shortfall is stated just as plainly. She "interacts reasonably effectively with her partner; she responds to what he says", but "she does not invite her partner to respond to what she says".

Then the part worth pausing on. Their Global Achievement marks are identical at 4.5. Better grammar, better discourse and better pronunciation bought Kenza exactly nothing in the overall mark, because asking her partner one question would have closed the only gap she had.

At B1 interaction decides the profile

Kenza is not a one off in the direction you would expect. She is the exception, and the rule runs the other way.

Across the nine published B1 candidates, Interactive Communication is the single highest mark for four of them and the highest or joint highest for eight. Pronunciation is the single highest for none.

The commentary for Andrea shows how little it takes. She scores 5 for Interactive Communication while the examiner is openly lukewarm: she "mostly responds to Cristina's contributions by agreeing with her", and "generally, however, she is more reactive than proactive".

She takes the top mark anyway, because she asks one question of her own and initiates on a couple of occasions with comments that invite a response. The commentary quotes one: "Also the water is too important for your health and if you get tired".

The same document notes that Andrea mispronounces a word, and her pronunciation mark is 4.5, level with her grammar and her discourse. A mispronounced word costs nothing visible. Sitting silent until spoken to costs half a band.

If you are preparing for this level, B1 Preliminary Part 3 and Part 4 are where that mark is won, because both are conversation with the other candidate. Part 1 and Part 2 give you far less room to initiate.

At C2 pronunciation stops carrying anyone

C2 Laura is the first candidate in the published set whose pronunciation mark sits below her discourse, interaction and global marks. She scores 2.5 for pronunciation against 3 for the other three.

Here is what earns a 2.5 at C2. She "is almost always intelligible". Her word stress "is almost always accurately placed". Her individual sounds "are almost always articulated clearly".

Every one of those phrases would read as praise at B1. At C2 they are the ceiling, not the floor. What pulls the mark down is stated separately: intonation "generally appropriate, but with little variation and often uneven, produced in short bursts that can be hard to follow", and "little variation in sentence stress".

Alejandro, her partner, scores 2.5 for pronunciation too, level with everything else in his profile except Global Achievement.

So at C2 the criterion behaves like any other. It can be your best mark, as it is for Jules, whose 4 for pronunciation is the only figure above his flat 3.5. It can equally be your worst. That is exactly what it never did at B2 or C1.

The three C2 Proficiency parts put different pressure on it. The long turn is where uneven intonation across a sustained stretch shows up most clearly, while the collaborative task tests it in short exchanges instead.

Why one criterion behaves differently at each level

The criterion name is constant. The standard behind it is not.

Cambridge assesses each candidate against the descriptors for the level they are sitting. At B1, intelligibility is the substance of the pronunciation criterion, and most candidates who reach a B1 exam room are intelligible. A criterion nearly everyone satisfies cannot separate people, so it clusters in the middle of the profile.

Interactive Communication at B1 has no such ceiling. Initiating, developing a partner's point and negotiating towards an outcome are strategy, and candidates vary enormously in whether they do them at all. So that column spreads, and it decides the profile.

At C2 the pronunciation standard has moved up to the deliberate use of phonological features to convey and enhance meaning. Being clear is assumed. The criterion has room to spread again, and it can land at the bottom.

B2 and C1 sit in the comfortable middle where being intelligible already reads as strong. That is precisely why nine candidates in a row scored well on it, and why we should not have expected that to travel.

This also explains something in the CEFR level descriptors that confuses learners: accent is never mentioned in them at any level, yet pronunciation is a scored criterion throughout. The descriptors describe what you can do, not how you sound doing it. Our guide to what C2 actually means goes through the same wording.

What this changes about how you practise

Practise in proportion to the level you are actually sitting.

At B1, put your time into initiating. Ask your partner a question, add a reason to their point, disagree occasionally. Mohammed's "what do you think about this bag" is the entire technique.
At B2 and C1, pronunciation is already the strongest column in every published profile. Extra accent drilling buys the least. Grammar and vocabulary range sits at the bottom of most of those profiles.
At C2, pronunciation is back in play, but not the part you expect. Sentence stress and intonation variation are what the commentary marks down, and neither improves by repeating single words.
At every level, the collaborative parts carry the interaction mark. B2 First Part 3 and C1 Advanced Part 3 are the same shape.

All four are conversation problems rather than study problems, which is why role play scenarios and a speaking partner you can talk to daily tend to move them faster than a textbook. If you are unsure which level you are working at, test your speaking level free first, because the advice above genuinely differs by level.

What we could not verify

This is a published sample, not a random one. Twenty three candidates across four levels. These are the candidates Cambridge chose to publish, selected to illustrate marking rather than sampled at random, so treat the proportions as illustrative.

We are not claiming causation. That Interactive Communication is the single highest mark in four B1 profiles is a fact about those profiles. Why it happens is our reading of the commentary, not a Cambridge statement.

Seven of the nine B1 candidates sat B1 Preliminary for Schools. Kenza and Mohammed sat B1 Preliminary. The criteria and marking are the same, and Cambridge lists both on the same preparation page, but the school candidates are teenagers discussing school trips and we cannot rule out that task content shapes which criterion spreads.

One candidate has a completely flat profile. Yaheli scores 4.5 on all five criteria, so she is simultaneously joint highest and joint lowest on every one. We have counted her as neither in the summary figures and flag her here rather than letting her quietly inflate a number.

We did not check the Linguaskill or Aptis criteria against these. Those tests report differently, as our Linguaskill speaking guide and Aptis speaking guide set out.

How we researched this guide

Every score in the table comes from a Cambridge English examiner commentary PDF, downloaded from the official preparation pages and read in full rather than summarised.

Two corrections happened during the research, and both changed the article, so both belong here.

The first was a counting error we caught before publishing. Our initial draft said B1 had four published candidates and that Interactive Communication was the top mark for all four. Opening the B1 Preliminary preparation page in a browser and reading the actual link list showed four commentary documents, not the three we had assumed, covering nine candidates rather than four. The extra five include Kenza, the one candidate in the whole set whose Interactive Communication mark is her lowest. A clean claim became a qualified one, which is what the evidence actually supports.

The second was a text extraction fault. The C2 score block renders with a stray space inside some numbers, so "2.5" extracts as "2 .5". A first pass read Alejandro as scoring 2 for Grammatical Resource and 2 for Interactive Communication, which would have made his profile look uneven. Pulling the complete score block rather than matching individual lines showed he scores 2.5 across the board with a 3 for Global Achievement.

Quotations are transcribed from the commentary text with the wording intact, including the hedges, because at this level of detail the hedges carry the meaning.

Practise speaking, from SpeakShark

SpeakShark is an AI English speaking practice app, and the reason it fits this particular finding is that the criterion doing the separating at B1 is interaction. You cannot drill interaction alone. It needs someone on the other side who says something you did not plan for, so that initiating and responding become real rather than rehearsed.

You talk, the AI answers what you actually said, and you get speaking feedback while the conversation is still running. The free tier gives basic feedback; the detailed pronunciation and grammar breakdown is on Premium.

Being straight about the limits: a free session runs five minutes with four turns, which is daily practice rather than a full test simulation. If you need to rehearse a complete Cambridge speaking test end to end with a partner beside you, this is not that.

Start free with three sessions a day and no card. Paid sessions run ten minutes with unlimited turns. Limits are on the pricing page, and how it works walks through a session. Sign up here if you want to practise asking questions back before your next class.

We are a speaking improvement tool. We are not an exam preparation provider, and we are not affiliated with Cambridge English, the IELTS partners, Pearson or any exam board. For test format, booking and official practice material, go to the exam body directly. Use SpeakShark to make your spoken English stronger, and use official material to learn the test.

Site: speakshark.com
Support: speaksharksupport@gmail.com
Free tier: 3 AI conversation sessions per day, no card

Sources

Cambridge English, B1 Preliminary preparation: the page that publishes all four B1 commentary documents, covering both B1 Preliminary and B1 Preliminary for Schools
Cambridge English, B1 examiner comments, Kenza and Mohammed: both score tables and every quotation in the Kenza and Mohammed section, including the invite a response comment and the moving the discussion forward comment
Cambridge English, B1 examiner comments, Cristina and Andrea: both score tables and the reactive rather than proactive verdict
Cambridge English, B1 examiner comments, Laura and Claudia and Yaheli, Fernanda and Lucia: the remaining five B1 score tables
Cambridge English, C2 Proficiency preparation, Laura and Alejandro and Manuel, Jules and Thiago: the complete published C2 set, all five score tables and every pronunciation quotation
Council of Europe, CEFR Table 1, common reference levels global scale: the level descriptors the criteria are applied against, none of which mention accent

FAQ

Are the Cambridge Speaking criteria the same at every level?

Almost, but not quite. B1 Preliminary and B2 First both use five criteria: Grammar and Vocabulary, Discourse Management, Pronunciation, Interactive Communication and Global Achievement. C1 Advanced and C2 Proficiency use six, because the combined Grammar and Vocabulary criterion splits into Grammatical Resource and Lexical Resource. The other four keep the same names at all four levels, which is what makes the published scores comparable across levels in the first place.

Which criterion decides a B1 Preliminary speaking profile?

Interactive Communication, on the published evidence. Across the nine B1 candidates whose scores Cambridge publishes with examiner commentary, it is the single highest mark for four of them and the highest or joint highest for eight of the nine. Pronunciation is the single highest mark for none of the nine. One candidate, Kenza, breaks the pattern in the other direction, and Interactive Communication is her single lowest mark.

Is pronunciation always a candidate's strongest criterion?

No, and this is where a widely repeated claim falls apart. Across the nine published B2 First and C1 Advanced candidates, pronunciation is the highest or joint highest mark every time. Widen the sample to all 23 published candidates from B1 to C2 and it is the single highest mark just three times, and it sits at the bottom of the profile for four candidates, all of them at B1 or C2.

What does a low pronunciation mark at C2 actually mean?

Not that the candidate is hard to understand. The C2 candidate scoring 2.5 for pronunciation is described as almost always intelligible, with word stress almost always accurately placed and individual sounds almost always articulated clearly. What costs her the mark is variation, specifically intonation produced in short bursts with little variation in sentence stress. At C2 the bar has moved from being understood to using phonological features with range.

Can a candidate with weaker English score higher on interaction?

Yes, and Cambridge publishes a case of it. Kenza and Mohammed sat the same B1 Preliminary test together. Kenza scores half a band above Mohammed for grammar and vocabulary, discourse management and pronunciation. Mohammed scores half a band above Kenza for Interactive Communication, and the examiner comment explains why: he asks his partner questions and moves the discussion forward. Their Global Achievement marks come out identical.

Where can I read these examiner commentaries myself?

Cambridge publishes them as free PDFs on the preparation pages for each qualification, next to videos of the Speaking tests they describe. Each document works through the criteria for every candidate in the test, quotes what the candidate actually said, then gives the score table. They are the closest thing available to watching your own test being marked, and they cost nothing.