Two different HSK word counts circulate because they come from two different documents. The 2021 Standards list 11,091 lines; the exam syllabus published in November 2025 has 11,000 entries and is the one the December 2026 exam follows. Below: the per-level counts for all three lists, the 1,949 old words that changed level, and the 875-word gap at HSK 4. Every row links to the page it came from.
Because a proficiency standard and an exam syllabus are different documents. The 2021 Standards (GF0025-2021) define what proficiency looks like at each level; the syllabus published in November 2025 lists the entries the exam actually tests. People quote both and end up arguing. Row two below is the standard, row three is the exam.
| List | Lines | Unique headwords | What it is |
|---|---|---|---|
| HSK 2.0 (2012) | 5,001 | 4,995 | The six-level exam that is ending |
| 2021 Standards (GF0025-2021) | 11,091 | 10,944 | Proficiency standard, not an exam syllabus |
| HSK 3.0 syllabus (Nov 2025) | 11,000 | 10,940 | What the exam from 13 Dec 2026 follows |
The 11,091 vs 11,000 difference is not a rounding error: the 2021 list repeats 147 headwords across levels, and the 2025 syllabus numbers every entry exactly once (ids 1–11,000). We counted both ourselves rather than quoting either figure.
The first official worldwide sitting is 13 December 2026, per the Chinese Testing International notice of 8 September 2026. Registration closes 16 November 2026 for paper and 3 December 2026 for computer. July 2026 is not a test date — it is the implementation date printed on the syllabus cover, and quoting it as a launch date is the single most common error in circulation.
Level 4 introduces 1,000 new entries and takes you to 2,000 cumulative; the jump to levels 7–9 accounts for 5,600 of the 11,000. Cumulative counts are what exam prep should plan against, not per-level counts.
| Level | New entries | Cumulative |
|---|---|---|
| HSK 1 | 300 | 300 |
| HSK 2 | 200 | 500 |
| HSK 3 | 500 | 1,000 |
| HSK 4 | 1,000 | 2,000 |
| HSK 5 | 1,600 | 3,600 |
| HSK 6 | 1,800 | 5,400 |
| HSK 7–9 | 5,600 | 11,000 |
Of the 5,001 old words, 1,949 moved up — 1,679 of them straight into the 7–9 band — 370 moved down, 126 were dropped, 2,519 stayed put and 37 are genuinely ambiguous. The movement is not confined to rare vocabulary: 饺子, 包子 and 面包 came down to level 1 while 小姐 went up to level 5.
| Word | Pinyin | Old → New | Syllabus page |
|---|---|---|---|
| 小姐 | xiǎojiě | L1 → L5 | 161 |
| 饺子 · 包子 · 面包 | jiǎozi · bāozi · miànbāo | L3–4 → L1 | 80–83 |
| 黑 · 红 | hēi · hóng | L2 → L5 | 142 |
| 鼻子 | bízi | L3 → L4 | 105 |
875 words. Old HSK 4 covered 1,200 cumulative words; new HSK 4 assumes 2,000. 1,125 of your existing 1,201 old-L1-to-L4 words are still in the new L1–L4 range, so the remainder plus new material is 875. Level 4 alone introduces 1,000 entries, 399 of which are entirely new word forms.
It is not published by the Center for Language Education and Cooperation or Chinese Testing International, and
it is not a textbook. 37 old words have multiple readings with no single defensible target, so they are marked
ambiguous instead of being assigned. 39 entries that share a character with an old word but have a
different reading are counted separately rather than folded into the 6,130. Where we could not decide, the row says
so — that is the difference between a page you can check and a list you have to trust.
网上流传两个 HSK 词数（11,000 与 11,091），两个都是真的，但来自不同文件：11,091 是 2021 年的 《国际中文教育中文水平等级标准》，11,000 是 2025 年 11 月发布的《HSK 考试大纲》——2026 年 12 月 13 日的考试 按后者执行。这份映射把旧版 5,001 个词逐条对上，每一行都标了官方 PDF 的物理页码，可以自己翻页核对； 37 个一词多音的词标为 ambiguous，没有替读者猜。
All 11,000 entries were extracted from the official syllabus PDF and verified by re-reading the page text: every entry's headword appears on the page recorded for it (100% of 11,000). Matching to the old six-level list is by headword plus normalised pinyin, with single-signal fallbacks only where the candidate is unique. Full CSVs and the extraction method: github.com/sadofriod/hsk-3-migration-map. Examples in the paid pack are drafted with AI assistance and reviewed by a human; this page and the free one-pager contain only counts and page numbers, every one of which can be checked against the PDF.