About
What this is, and where the words come from
A small study tool for the HSK vocabulary that takes effect in December 2026. No accounts, no tracking, no server: every page is a static file.
The syllabus ladder
The tests
Each deck of 50 words ends in a 34-question test, and each level in a 60-question test. Both are marked on the spot, both show every answer afterwards, and both keep the result on your device.
A deck test is 20 recognition questions (the word, four meanings, five seconds), 4 production questions (the meaning, then its pinyin, twenty seconds), 6 sentence questions (fill the gap or put the words in order, sixty seconds) and 4 listening questions (a sentence is read aloud, then you choose it from three written lines, thirty seconds). It passes at 80%, and only when each section clears its floor: 70% recognition, 50% production, 50% sentences, 50% listening. The floors stop a strong recognition score from carrying a weak sentence or listening section.
The five seconds are there so the answer is what you remember rather than what you can look up. The clock can be doubled or switched off before the first question, and it stops while the tab is hidden.
A level test draws on every word in that level: 40 recognition, 5 production, 10 sentences and 5 listening, passing at 60% with floors of 60%, 40%, 40% and 40%. The sentence questions use the two formats that have appeared in HSK papers for years, 选词填空 (choose the word for the gap) and 排列顺序 (put the sentence in order). A retake draws new words, so it is never the same paper twice.
The listening questions read each sentence twice, first in a female voice and then in a male one, slowly, and again whenever you press the button. A browser without a Chinese voice cannot read the sentence, so there the question is put the other way round: the English is shown and you choose the Chinese line that means it.
The games
Five games sit beside the decks, on the 500 words of HSK 1 and 2: name the tone you hear, pair characters with meanings, pick the word you heard, choose an opposite, and put a sentence back in order. Each round takes under two minutes, and every word is read twice, female then male.
The rounds borrow from research on retrieval practice, spaced practice and tone training. That does not make them proven here. There has been no pre-test, no control group and no delayed test, so no learning gain is claimed. Match Pairs leans on retrieval practice; Opposite Duel is variety, not a core routine. The games page says what each round asks of you.
Dates
The syllabus was published in November 2025. The first official sitting in the new format is 13 December 2026. From January 2027 every sitting uses it. The 2021 standard, with its 1,110 syllables and its 300-word HSK 3, is being replaced. The syllabus publishes word lists only, which is why this site has no syllable chart.
Sources
The word lists come from the syllabus published in November 2025. The traditional forms and English meanings are adapted from CC-CEDICT and published by Second Language under CC BY-SA 4.0. Part-of-speech labels are the syllabus's own. This site is unofficial and is not affiliated with Chinese Testing International.
CC-CEDICT glosses are written for a dictionary, not for a beginner. Where a word has several senses, the list shows the common ones separated by semicolons. If a gloss does not match the sense you were taught, the dictionary is the likely source of the difference.
The example sentences come from open sets that allow reuse: the HSK sentence collection published as no7z/hsk-sentences-audio, the Gushi HSK 3.0 lists and Tatoeba, credited below. A sentence was kept only when every word in it sits at or below the level of its card, and the pair on each card was then read and corrected by hand. Synonyms are drawn from the Chinese Open WordNet and 同义词词林扩展版 (HIT Information Retrieval Lab), and opposites from a hand-checked pair list. Where no usable synonym or opposite exists, the card carries none rather than a wrong one.
Colour
Blue is the field, and white is the only text on it; a tint toward blue or red costs more contrast than it buys, so size and weight carry the hierarchy instead. Cinnabar red is the card, so a word sits on a seal rather than on a sheet. A correct answer in the quiz turns white, and a wrong one sinks under a dark wash and carries a cross, so colour is never the whole message.
- Mehta & Zhu, Science 323 (2009): blue improves creative performance, red improves detail-oriented performance.
- Eto et al., JSPS KAKENHI 16K01076: background colour and screen reading accuracy.
- Kaplan, attention restoration theory: natural colour restores directed attention.
- Buchner & Baumgartner (2007): dark text on a light background reads better on screen. This palette goes the other way on purpose. White on the blue field measures 3.83:1, which clears the 3:1 bar for large type and falls short of the 4.5:1 one for body copy.
- Elliott & Maier, Annual Review of Psychology 65 (2014): colour effects depend on context and saturation.
Those figures are computed from the colours actually rendered, not estimated. White on the seal card measures 6.45:1, and the button label on its white face measures 5.57:1. The field is #007FFF and the card #AB342B. Light and dark follow your system setting; there is no theme switch.
What this site stores
Which decks you have opened is kept in your browser's local storage under the key
hsk2026_opened_v1. Test results go to hsk2026_tests_v1, one record
for each deck test or level test holding the mark, the date and whether the clock ran, and the
clock choice is kept in hsk2026_timing_v1. Drawing and speaking practice keeps
an average for each of those two modes under hsk2026_practice_v1: a count and a
total per deck, never the microphone audio or the transcript. Game bests sit under
lab_best_tone, lab_best_match, lab_best_listen,
lab_best_opposite and lab_best_forge. None of it is sent anywhere. There are no
cookies, no analytics and no accounts. Clearing site data resets your progress, your marks and
your practice scores.
The privacy policy sets out the rest, including the other parties your
browser talks to when you use speech recognition.
Attribution
Unofficial study aid, not affiliated with Chinese Testing International. Word data from the HSK 2026 syllabus and CC-CEDICT, shared under CC BY-SA 4.0.
Example sentences: the no7z HSK sentence set and the Gushi HSK 3.0 lists, both CC BY-SA 4.0, and Tatoeba, CC BY 2.0 FR. Synonyms: the Chinese Open WordNet and 同义词词林扩展版, HIT Information Retrieval Lab. Sentences reused under CC BY-SA 4.0 keep that licence here.
Character drawing: Hanzi Writer, MIT licence. Stroke outlines and medians come from Hanzi Writer Data, Arphic Public License, derived from Make Me a Hanzi. The licence texts ship with the site: Hanzi Writer and Arphic.