How we measure

Last updated: November 8, 2026 · by Kerem Yiğit Elbasan

The word list

For each language we start from word frequency data (the wordfreq project, which combines subtitles, web text, books and other sources) and group inflected forms under one dictionary word: “compré” counts as “comprar”. We keep the 8,000 most frequent dictionary words. Each one has a short meaning, examples and links to the forms that point to it.

The vocabulary test

The 8,000 words are split into 10 frequency bands: 0–400, 400–800, 800–1,200, 1,200–1,800, 1,800–2,500, 2,500–3,200, 3,200–4,000, 4,000–5,000, 5,000–6,500 and 6,500–8,000. The test asks 2 random words from each band, 20 in total, with 4 choices and an “I don’t know” button.

For each band we take (right − wrong ÷ 3) ÷ questions and multiply by the band’s width. The penalty for wrong answers corrects for lucky guesses: with 4 choices, a pure guesser gets one right for every three wrong. We add the bands and only then limit the total to 0–8,000.

Why only at the end: an earlier version limited each band to 0 separately, which quietly rewarded guessing. In a simulation of people answering completely at random, that version gave about 1,500 words on average; the current one gives about 470. We changed it in November 2026.

Two questions per band is a small sample. A single answer can move the result by several hundred words in the wide bands, which is why the result is shown as a range, not a single number, and why the test says “rough estimate”.

Video levels

When a video is added, we read its subtitles, map every word to its dictionary word and find the smallest vocabulary from our list that covers 95% of the words spoken. That number decides the CEFR-style level shown on the video. Names in capital letters are not counted as unknown, and in languages other than English an unknown word counts at most three times so one repeated term does not push a video up a level.

What we have measured so far

Across the 247 Spanish videos in the app (November 2026, all levels), someone who knows the 800 most common words recognises on average about 82% of the words they hear, 2,000 words about 90%, and 8,000 words about 97%. The 385 English videos give almost the same curve (81%, 89%, 97%).

These are averages measured with our own word list. Learner channels sit higher, native content lower, so treat the numbers as an illustration, not a promise.

Limits

A word count is not fluency. Knowing a word in a list is not the same as catching it at full speed in a noisy video. Automatic subtitles contain errors, and frequency lists reflect the sources they were built from. We publish these numbers so you can judge them, and we will update this page when the method changes.

Try the Spanish word test