HSK 2.0 vs 3.0 Word-List Differences
Same HSK digit, different word list. The 2009 totals sit beside the 2025 syllabus, with a pinned third-party file for what an entry-level diff can claim.
By Chinese CopilotLast reviewed ≈11 min read
On this page
The same digit on an Hanyu Shuiping Kaoshi (HSK) chart is not the same word list. Level 1 in the 2009 introduction is a vocabulary total of 150. Level 1 in the 2025 syllabus is sequence number 300. Level 4 is 1,200 on that 2009 page (the page prints 1200) and number 2,000 on the syllabus. Level 6 is Over 5000 on the introduction and number 5,400 on the syllabus. Levels 7-9 have no row on the 2009 page. On the syllabus they are one slice, numbers 5,401 to 11,000, which is 5,600 new rows.
The introduction, titled "Introduction on New HSK Test" and dated 2009-10-14, prints those six vocabulary totals and says the test will be launched in November 2009. It does not print the words. The 2025 syllabus PDF, 《中文水平考试 HSK 考试大纲》 (Syllabus for the Chinese Proficiency Test), issued by the Center for Language Education and Cooperation (CLEC), prints one list numbered 1 to 11,000. The cover prints 2025-11 发布 and 2026-07 实施. Cite-line totals live in HSK 3.0 by the Numbers. Stage names live in HSK 2.0 vs HSK 3.0: What's Changing?. The three-column size table, including the 2021 standard GF 0025-2021, lives in Chinese Vocabulary Size vs HSK Level. This page leaves that column there.
An entry-level diff needs a named file. The comparison below uses complete-hsk-vocabulary at commit 7ac65bf1a6387d35f1ade478906172a19311c7f9 (23 March 2026, message "Auto-generate vocabulary lists"). The license is MIT, copyright Yanis Zafirópulos (Dr.Kameleon). It is a third-party file. It is not an official list from the 2009 introduction or from CLEC.
2009 vocabulary column beside 2025 sequence numbers. Levels 7-9 exist only on the later list.
Two official totals, two documents
| Level | 2009 vocabulary total | 2025 last number | 2025 new rows | Open it |
|---|---|---|---|---|
| 1 | 150 | 300 | 300 | Level 1, list |
| 2 | 300 | 500 | 200 | Level 2, list |
| 3 | 600 | 1,000 | 500 | Level 3, list |
| 4 | 1,200 | 2,000 | 1,000 | Level 4, list |
| 5 | 2,500 | 3,600 | 1,600 | Level 5, list |
| 6 | Over 5000 | 5,400 | 1,800 | Level 6, list |
| 7-9 | no row on that page | 11,000 | 5,600 | What Is HSK 7-9?, list |
The 2009 column is a cumulative vocabulary total. Neighboring cells differ by 150, 300, 600, and 1,300. Level 1's cell is the total 150, not a second gap of 150, and Level 6 then stops at the words Over 5000. Those differences are arithmetic. The page does not publish them as a new-word list. The 2025 column is the last sequence number on one list. Parenthetical tags such as 1（4） stay inside the earlier numbers. Chinese characters are a separate outline: Chinese Characters vs Words. Sitting dates are in Should You Take HSK 2.0 or HSK 3.0 in 2026?. Section counts are in HSK 3.0 Exam Format.
Stripping a trailing digit from words such as 本1 leaves 10,896 distinct forms across the 11,000 rows, so 104 further rows belong to a form already counted. The syllabus does not print 10,896 as a heading. HSKStory publishes that endpoint and a Level 2 cumulative of 496. Counting each form at its earliest number on this extract gives 300, 497, 988, 1,978, 3,557, 5,334, and 10,896.
Syllabus rows on one side, a pinned third-party file on the other. The file's new-level tags are a separate column.
What a careful diff can claim
The 3.0 side of this page is the 2025 syllabus. The word-by-word 2.0 side is only the old-1 through old-6 tags in that pinned file. The README calls old-N HSK 2.0 and new-N, including new-7, HSK 3.0. Those are the file's names.
A match means the same simplified characters. A trailing ASCII digit is removed first, so 本1 and 本2 both meet the single file entry 本. The join is a written form. It is not a sense match, and it is not a pinyin match. 不必 is number 1043, Level 4, búbì, while the file's reading is bù bì. 聪明 is number 570, Level 3, cōngmíng, while the file's reading is cōng ming.
The file has 11,470 entries. Exactly 4,991 carry one distinct old-* tag: 150, 147, 298, 598, 1,298, and 2,500. No entry carries two different old levels. Seven repeat the same tag, including 还 stored as old-2 twice. Those counts sit near the gaps in the 2009 column, and they are not those gaps: the file has 147, 298, 598, and 1,298 where a subtraction of the printed totals would be 150, 300, 600, and 1,300. Level 6 on the page is Over 5000, so it does not yield a fixed new-word integer. The file's old-6 count is 2,500.
Of those 4,991 forms, 4,802 appear on the syllabus and 189 do not. Missing by old level: 3, 6, 11, 13, 28, and 128. A missing form is absent from this comparison. It is not a deletion notice from CLEC. The three old-1 forms with no syllabus row are 北京, 火车站, and 没. 没有 is a different row, number 122, Level 1, méiyǒu, and that file entry has no old-* tag.
The last column counts forms whose earliest syllabus row is in that level. It is a count, not a sequence number. Fifty-nine old forms also have a later syllabus row, and that later row stays out of the count. 还 is number 67, Level 1, hái, and number 678, Level 3, huán. The file has one 还. The earliest row is Level 1.
| File tag | Entries | On the syllabus | Missing | Earliest row in that level |
|---|---|---|---|---|
| old-1 | 150 | 147 | 3 | 143 in Level 1 |
| old-2 | 147 | 141 | 6 | 70 in Level 2 |
| old-3 | 298 | 287 | 11 | 226 in Level 3 |
| old-4 | 598 | 585 | 13 | 463 in Level 4 |
| old-5 | 1,298 | 1,270 | 28 | 768 in Level 5 |
| old-6 | 2,500 | 2,372 | 128 | 679 in Level 6 |
| Total | 4,991 | 4,802 | 189 |
old-2 also lands 55 of its on-syllabus forms in Level 1. Syllabus forms with no old-* tag, counted at the earliest row, are 91, 83, 186, 381, 599, 840, and 3,914, summing to 6,094. Of those, 5,255 are in the third-party file with no old tag, and 839 have no entry in the file. "Absent from this file" fits only the 839.
The file's new-* tags are another column: 506, 750, 953, 972, 1,059, 1,123, and 5,606, summing to 10,969. Of those forms, 9,696 appear on the syllabus and 1,273 do not. newest-* tags cover 10,057 forms, all of them on the syllabus, while 839 syllabus forms have no newest-* tag. The syllabus sequence numbers stay the official column.
Digit 1 is 150 in 2009 and number 300 in 2025. 小姐 is one syllabus row, number 3269 at Level 5, and old-1 only in the pinned file.
What other comparison pages leave unnamed
On 20 October 2026 three public comparisons were already up. None of them pinned this commit.
Saymei, updated 8 October 2026, says: "We matched every one of the 4,992 distinct HSK 2.0 words against the new syllabus. Of those, 4,856, or 97.3%, are still on the HSK 3.0 list." A table note says "Source: our comparison of the two official lists." The 2009 introduction prints no words, so that note does not name a second official list. The page names no commit. Its method note parses 11,000 entries (10,896 distinct forms), describes a six-level list of 5,000 entries and 4,992 distinct forms, and cites 11,092 words for the 2021 standard. This page does not re-read that standard. A kept word is the same written form, with erhua variants accepted, so their 136 and this file's 189 are different counts. This page does not adopt 4,992 or 97.3%.
HSKStory labels 300, 496, 988, 1,978, 3,557, 5,334, and 10,896 as repository output beside the official line 300, 500, 1,000, 2,000, 3,600, 5,400, and 11,000. The official line matches the syllabus. The normalized Level 2 figure is 496, one below this extract.
ReadChinese says "The 2021 HSK 3.0 reform significantly restructured the vocabulary list" and displays 4,991 and 10,969. Its columns are the same old-* and new-* counts listed above, from 150 to 2,500 and from 506 to 5,606. The page names no commit, so 10,969 is easy to mistake for the syllabus. The syllabus new-row counts are 300, 200, 500, 1,000, 1,600, 1,800, and 5,600.
Saymei, ReadChinese, and HSKStory name no commit. The lower row is this pinned file, set beside syllabus number 500.
Rows you can open
The list URL plus the syllabus number is the check. The index is HSK 3.0 vocabulary. On Levels 5, 6, and 7-9, a letter heading marks a run of rows that share a pinyin initial. The same letter can return later, because those rows stay in syllabus order.
爱 is number 1, ài, gloss "to love; to be fond of; to like", tagged old-1. 饺子 is number 90, jiǎozi, gloss "dumpling", tagged old-4. Saymei also says 4 then 1. That 2.0 cell is still third-party. 包子 is number 8, bāozi, gloss "bao (steamed stuffed bun)", tagged old-5, while Saymei says 2.0 level 4. 元 is number 268, tagged old-2, while Saymei says 2.0 level 3. 视频 is number 1673, shìpín, gloss "video", with no old-* tag, while Saymei says 2.0 level 6. The 3.0 cells match the syllabus. Saymei does not name the 2.0 file.
女士 is number 155, tagged old-5. 小姐 is number 3269, xiǎojiě, tagged old-1. Saymei says 5 then 1, and 1 then 5. Open row 3269 for the English gloss. 红 2483, 白 2012, and 黑 2482 are Level 5 and old-2.
半 is number 7, level 1（4）, gloss "half; semi-; incomplete", tagged old-3. The （4） stays on Level 1. That parenthetical tag also covers 对 44, 叫 91, 日 172, 一下 259, and 真 285. 一下 has no entry in the third-party file. 分 is number 55, level 1（3）, tagged old-3. 正 is number 491, level 2（5）, tagged old-5.
还 number 67 is hái, and 还 number 678 is huán. One file entry carries old-2 twice. 本1 is number 10 and 本2 is number 2040, against one 本 tagged old-1. 啊 is number 301 and 啊 is number 1001, against one 啊 tagged old-3.
扫码 is number 1627, gloss "to scan a QR code or barcode", with no file entry. 爱好 is number 302, tagged old-3. Level 2 starts at 301. 做饭 is number 300, with no old tag. 岸 is number 3601, tagged old-5. 罪 is number 5400, with no old tag.
One slice, 5,401 to 11,000. The 1,525, 679, and 168 counts are earliest rows from the pinned file.
Levels 7-9, and how the lists are built
The syllabus does not split 7, 8, and 9 into three vocabulary lists. Of the 2,372 third-party old-6 forms on the syllabus, the earliest row is inside Levels 7-9 for 1,525, inside Level 6 for 679, inside Level 5 for 155, and inside Level 4 for 13. Those four counts add to 2,372. They are not a deletion list from CLEC. Number 5,401 is 挨. Number 11,000 is 做主, zuòzhǔ, part of speech 动. The English gloss is empty, so the page shows a dash. The file tags 做主 old-6 and newest-7. The other endpoints are 爱 to 做饭, 啊 to 左边 (301 to 500), 阿姨 to 作业, 啊 to 作者, 哎 to 作出, and 岸 to 罪.
Words, numbers, levels, and pinyin come from the 2025 syllabus. English glosses are adapted from CC-CEDICT, including the MDBG snapshot dated 2026-10-03T08:59:08Z, under CC BY-SA 4.0, based on CEDICT © 1997, 1998 Paul Andrew Denisowski. Some glosses are shortened, and some common words are hand-corrected, still under that license. The commit above processed the dictionary. MIT covers the processing, not the meanings. An empty gloss renders as a dash. Grouping terms for the checks are in the methodology glossary, and the full account is on the methodology page.
119 words, 17 from each group, a fresh draw each run. Then open the list and read the number.
The HSK 3.0 test is a recognition check. Each run draws 17 words from each of Levels 1 to 6 and from the combined 7-9 list, 119 words in all, and the next run draws again. The card shows characters and pinyin. The result names an approximate level from the words you say you recognize. It is free, it needs no account, and it is not an official exam result. Numbered homographs stay on the list pages and stay out of the draw. After a run, open the list for the level the estimate named and read the rows. The digit alone does not tell you which list you are in.
More on the new exam, level by level, is in our HSK 3.0 guide.
Sources
- Introduction on New HSK Test, chinesetest.cn, 2009-10-14. Vocabulary totals 150, 300, 600, 1200, 2500, and Over 5000. No word-by-word list.
- 2025 HSK exam syllabus PDF, CLEC. One numbered vocabulary list, 1 to 11,000. Cover: 2025-11 发布, 2026-07 实施.
- Hanyu Shuiping Kaoshi, Pinyin, and Chinese characters, English Wikipedia. Secondary overviews only. No figures in this post come from them.
- complete-hsk-vocabulary at commit 7ac65bf1a6387d35f1ade478906172a19311c7f9, 23 March 2026, MIT, Yanis Zafirópulos. Third-party
old-*,new-*, andnewest-*tags. Not an official list. - CC-CEDICT and the CC BY-SA 4.0 license. English glosses on the list pages. The MIT processing license does not cover those meanings.
- Saymei, "HSK 3.0 explained", updated 8 October 2026. Third-party overlap claim (4,992 and 97.3%) and example table. No commit named.
- HSKStory, "HSK 3.0 vs HSK 2.0". Official cumulative line and their normalized distinct-form line, labeled as repository output.
- ReadChinese HSK comparison. Third-party columns equal to this commit's
old-*andnew-*counts, framed as the 2021 reform.