Meet Felix — now for Speaking
Ask an ordinary chatbot how your answer sounded and it will tell you it was great. Felix does the opposite: he marks what you actually said on the Cambridge speaking scales, tells you whether it would be around a pass, and shows the exact words to swap for stronger ones.
Two exercises in every Speaking part are free to practise, no account needed. Felix’s marking comes with the passes that include AI feedback — compare the passes.
Every Speaking question gives you the same box: press the mic and answer out loud, and your words appear as you say them. No microphone, or somewhere you can’t speak? Type the answer instead — the drill finishes either way, and so does the marking.
Anything the recogniser mishears you can fix before you submit, and it wouldn’t be marked against you anyway.
The writing Felix corrects your text line by line. The speaking Felix can’t — and shouldn’t. You were talking, so he judges the things that matter when you talk.
You speak; the browser writes it down. Felix marks that transcript — how you actually talked, not a tidy essay you had time to polish.
The text came from speech-to-text, so odd capitalisation, missing full stops and the occasional misheard word are ignored. He judges the language and the ideas, never typos.
There's no audio sent for marking, so Felix can't and won't judge your accent or sounds. Record yourself and listen back for that — or ask a teacher.
B2 and C1 are held to different standards. A B2-level turn won't be handed a C1 pass — the band lands where a real examiner would put it.
A real Speaking test is marked on four scales; the fourth is Pronunciation, which needs your voice. From a transcript, Felix marks the three he honestly can.
How wide and how accurate your grammar and vocabulary are as you speak.
Range and accuracy of the grammar and vocabulary you used.
Speaking at length with clear, relevant, well-organised ideas — not drying up.
Speaking at length with clear, relevant, well-organised ideas.
Starting, responding and keeping the conversation going in Parts 1, 3 and 4.
Starting, responding and keeping the conversation going (Parts 3 & 4).
Each part asks for something different, so each part is checked against its own rules before the language is judged at all.
Did you answer naturally, with reasons and examples, using the right tenses?
Did you compare the photos (not just describe them), speculate, and answer the question for the whole minute?
Did you give opinions on the ideas, interact with your partner, and work towards a decision?
Did you give opinions backed by reasons and examples, and develop your answers?
“Use more advanced vocabulary” helps nobody. Every result ends on a short list of things you actually said, each with a stronger version and one line saying why it is better — so you have something to practise before the next attempt.
Every result says plainly whether the answer is around a pass at your level, gives a band for each scale, and lists concrete words to upgrade and things to fix. If you barely spoke, he says so and marks it low — no wall of compliments. That’s the point: you find out here, while you can still practise, not on exam day.
Pick a part, answer it out loud, and Felix shows you exactly where you stand — and exactly what to say differently next time.
Felix gives an AI estimate to guide your practice — a genuinely rigorous one, but not an official Cambridge result. Only a real examiner can award your final grade.
