OpenJLPT

The JLPT N5–N1 word, kanji and grammar lists: cleaned, cross-checked against JMdict, and free to use.

7,811 words · 2,383 kanji · 526 grammar points · example sentences with furigana for 93% of words

Learners: 🌐 Browse online · 🃏 Anki decks · 📖 Yomitan dictionary · 🔍 What level is this text?
Developers: npm i openjlpt · pip install openjlpt · JSON / CSV / SQLite · API · 日本語

Note

The JLPT hasn't published official vocabulary, kanji or grammar lists since 2010. Word and kanji levels here follow Jonathan Waller's community lists, the ones Jisho.org uses. Treat every level as a well-informed approximation. More on levels ↓

Why OpenJLPT?

Most JLPT datasets are copies of the same lists, and they come with the same garbled readings, no example sentences, and unclear licensing. OpenJLPT fixes the data and makes it easy to use:

• Verified against JMdict. 99.7% of words are linked to their dictionary entry, which adds part of speech and a JMdict ID and catches wrong readings. Hundreds of source errors are fixed, each backed by a unit test.
• Real example sentences, with furigana. Tatoeba sentences for 93% of words are matched by dictionary form, so 読む finds 読んでいる. Short, checked sentences come first, and about four in five carry readings over every kanji.
• 526 grammar points. Original, reviewed explanations with formation rules, 2–3 examples each, and notes on easily confused patterns.
• Ready for your stack. JSON, CSV, SQLite with full-text search, Anki, Yomitan, npm, PyPI, a CLI, and a CDN. Every entry has a stable ID, so saved progress survives data updates.
• Honestly licensed and kept fresh. CC BY-SA 4.0 with per-field attribution, rebuilt monthly from upstream, and schema-validated in CI.

What's inside

Level	Words	Kanji	Grammar
N5	674	84	81	Beginner
N4	630	169	98	Elementary
N3	1,659	387	101	Intermediate
N2	1,778	398	123	Upper intermediate
N1	3,070	1,345	123	Advanced
Total	7,811	2,383	526

Each word appears once, at the easiest level that lists it. The counts come from data/json/meta.json, which also records the upstream versions.

Quick start

No code

• The website has a page for every word, kanji and grammar point, plus search, spaced-repetition flashcards, and a "what JLPT level is this text?" analyzer.
• Anki decks for each level, with furigana, example sentences and text-to-speech.
• Yomitan dictionary that shows JLPT levels in your pop-up dictionary.
• Spreadsheets: open any CSV in Excel or Google Sheets.

JavaScript / TypeScript

npm install openjlpt

import { findWord, findKanji, searchVocab, getGrammar, kanjiIn } from 'openjlpt';

findWord('食べる');             // { reading: 'たべる', romaji: 'taberu', pos: ['v1', 'vt'], examples: [...] }
findKanji('日');                // { strokes: 4, radical: '日', onyomi: ['ニチ','ジツ'], words: ['明日', ...] }
searchVocab('taberu');          // kanji, kana, romaji or English; ranked
getGrammar('N4');               // 98 grammar points with examples
kanjiIn('日本語を勉強する').map((k) => k.level);   // ['N5', 'N5', 'N5', 'N4', 'N4']

In the browser or a bundler (Vite, Next.js, React Native): the main entry reads files with node:fs, so import the JSON directly instead, or fetch it from the CDN:

import n5 from 'openjlpt/data/json/vocab/n5.json';   // bundled, fully static
const n4 = await fetch('https://cdn.jsdelivr.net/gh/evanclan/OpenJLPT@main/data/json/vocab/n4.json').then((r) => r.json());

Python

pip install openjlpt

from openjlpt import find_word, search_vocab, get_grammar, query

find_word("食べる").meanings                  # ['to eat']
search_vocab("weather", level="N5")[0].word  # '天気'
query("SELECT word, reading FROM vocab WHERE level = 'N5' LIMIT 2")
# [{'word': '会う', 'reading': 'あう'}, {'word': '青', 'reading': 'あお'}]

Command line

$ npx openjlpt 食べる          # or: pip install openjlpt && openjlpt 食べる
[N5] 食べる たべる taberu
   to eat
   Ichidan verb, transitive verb
   › ちょうど食べたかったものでした。
     That hit the spot.

$ npx openjlpt kanji 日        # kanji details
$ npx openjlpt grammar ながら   # grammar points
$ npx openjlpt quiz N4         # 10-word reading quiz in your terminal

SQLite

-- Full-text search over words, readings, romaji and meanings
SELECT v.word, v.reading, v.level
FROM vocab_fts f JOIN vocab v ON v.rowid = f.rowid
WHERE vocab_fts MATCH 'weather';

API at a glance

TypeScript	Python	Returns
getVocab(level?)	get_vocab(level=None)	all words, or one level's
findWord(word) · findWords(word)	find_word(word) · find_words(word)	a word by spelling (also alternative spellings) · all homographs
getVocabById(id)	get_vocab_by_id(id)	a word by its stable ID
searchVocab(q, {level, limit})	search_vocab(q, level, limit)	ranked search over kanji, kana, romaji and English
getKanji(level?) · findKanji(ch)	get_kanji(level) · find_kanji(ch)	kanji
kanjiIn(text)	kanji_in(text)	the JLPT kanji in a text, in order
getGrammar(level?) · getGrammarById(id)	get_grammar(level) · get_grammar_by_id(id)	grammar points
findGrammar(pattern) · searchGrammar(q)	find_grammar(pattern) · search_grammar(q)	grammar lookup and search
meta() · posLabels()	meta() · pos_labels()	counts, source versions · part-of-speech legend
sample(items, n)	sample(items, n, rng)	random picks for flashcards and quizzes
toRubyHtml(furigana)	to_ruby_html(furigana)	an example's furigana as <ruby> HTML
—	query(sql, params) · connect()	the bundled SQLite database (read-only)

Download

Format	Where
JSON, per level	data/json/: vocab/, kanji/, grammar/ × n5.json … n1.json
CSV, per level	data/csv/
SQLite, everything	data/openjlpt.sqlite: tables vocab, kanji, grammar, pos and the FTS index vocab_fts
Everything in one zip	openjlpt-data.zip from the latest release: JSON, CSV and SQLite
pandas, R, spreadsheets	pd.read_csv("https://cdn.jsdelivr.net/gh/evanclan/OpenJLPT@main/data/csv/vocab-n5.csv")
Anki decks, Yomitan	the website and Releases

Data format

Every entry is validated against the JSON Schemas in CI.

Vocabulary: data/json/vocab/n5.json

{
  "id": "86f67e2dc3",
  "word": "食べる",
  "reading": "たべる",
  "romaji": "taberu",
  "meanings": ["to eat"],
  "level": "N5",
  "pos": ["v1", "vt"],
  "jmdict_id": 1358280,
  "examples": [
    {
      "ja": "ちょうど食べたかったものでした。",
      "furigana": "ちょうど{食|た}べたかったものでした。",
      "en": "That hit the spot.",
      "tatoeba_id": 202896
    }
  ]
}

Field
id	Stable ID. It is assigned once and frozen in sources/ids.lock.json, so it survives spelling and reading fixes.
word · reading · romaji	Headword, kana reading (kana-only words read as themselves), and Hepburn romaji without macrons (toukyou).
meanings	English glosses, most important first.
pos · jmdict_id	JMdict part-of-speech codes for the matching sense (legend: v1 means "Ichidan verb"), and the JMdict entry number.
other_forms · other_readings	Other spellings (あさって → 明後日) and readings (四 → よん). Optional.
examples	Up to two Tatoeba sentences, with tatoeba_id for attribution. Optional.
examples[].furigana	The sentence with readings over its kanji, in {漢字|かんじ} notation. Present only when every kanji could be read with confidence. toRubyHtml() / to_ruby_html() turn it into <ruby> HTML.

Kanji: data/json/kanji/n5.json

{
  "character": "日", "level": "N5", "strokes": 4, "grade": 1, "freq": 1,
  "radical": "日", "radical_number": 72,
  "onyomi": ["ニチ", "ジツ"], "kunyomi": ["ひ", "-び", "-か"], "nanori": ["あ", "あき", "いる"],
  "meanings": ["day", "sun", "Japan", "counter for days"],
  "words": ["明日", "一日", "五日", "昨日", "今日", "九日", "十日", "七日"]
}

• freq is the newspaper frequency rank (1 = most common).
• radical is the classical (Kangxi) radical.
• kunyomi uses KANJIDIC2 notation: た.べる marks okurigana, - marks prefix or suffix use.
• words lists OpenJLPT words that use the kanji, easiest first.
• "supplementary": true marks the 172 jōyō kanji that are missing from Waller's pre-2010 lists (誰, 頃, and even 分). They are placed at the easiest level of a word that uses them, otherwise N1.

Grammar: data/json/grammar/n5.json

{
  "id": "te-mo-ii",
  "pattern": "〜てもいい",
  "romaji": "te mo ii",
  "level": "N5",
  "meaning": "may; it's okay to",
  "formation": "Verb-te + もいい",
  "examples": [
    { "ja": "この辞書を使ってもいいですか。", "en": "May I use this dictionary?" },
    { "ja": "今日は早く帰ってもいいです。", "en": "You may go home early today." }
  ],
  "tags": ["permission"],
  "notes": "Add ですか to ask for permission. A polite refusal is often just すみません、ちょっと…."
}

Patterns with kanji also have a kana reading (〜に対して → 〜にたいして). tags come from a controlled vocabulary.

How the data is built

Waller's lists, JMdict, KANJIDIC2 and Tatoeba go in, and scripts/build-*.ts turns them into data/. CI then validates the result. The level lists are the community standard, but the raw files are messy. The build fixes them systematically, and tests/ts/normalize.test.ts quotes the real source card behind every rule:

Problem in the source	Example	Fix
Missing readings on kana words	あさって → ""	reading = word (1,096 words)
する baked into readings	勉強 → べんきょうする	strip it (30 words)
Notes and mojibake as readings	はい → （感）, 賛成 → Uӣ[い	reject; fill from JMdict
Glosses split mid-parenthesis	to take (e.g. time + money)	split outside parentheses only
Glosses cut off at 100 characters	…;unwise;untime	repair from JMdict (untimely)
Wrong readings	途中 → つちゅう, 灰皿 → はいさら	corrected when kanji and meaning agree with JMdict
Typos in kanji	著 for 着, 田ぼ for 田んぼ	matched by reading and meaning
The same word at several levels	あさって (N5) and 明後日 (N3)	kept once, at the easiest level
Example sentences by substring	あれ matched 冷徹であれ！	matched by dictionary form via Tatoeba's word index
Jōyō kanji missing from the lists	分, 誰, 頃	added as supplementary

To rebuild everything yourself: npm install && npm run build. See CONTRIBUTING.md.

Alternatives: when to use something else

OpenJLPT stands on the shoulders of these projects. Pick the one that fits your need:

• jmdict-simplified: the whole JMdict and KANJIDIC2 as JSON. Use it if you need a full dictionary rather than JLPT lists.
• open-anki-jlpt-decks: JLPT vocabulary Anki decks. Use it if you only want Anki vocab decks in its format.
• kanji-data: kanji with JLPT and WaniKani levels. Use it if you need WaniKani levels.
• OpenJLPT fits best when you want vocabulary, kanji and grammar together, cleaned and linked to JMdict, with example sentences, in many formats, under one license.

FAQ

Can I use it in a commercial or closed-source app? Yes. The data is under CC BY-SA 4.0, the same license as JMdict, which many commercial apps already use. Credit OpenJLPT and its sources: one line on your About screen is enough (copy it from NOTICE.md). ShareAlike applies to the data you redistribute, such as a modified word list. (This is not legal advice.)

Why is word X at level Y? Levels come from Waller's lists. If one looks wrong, open an issue with a source. Wrong readings, meanings or examples are bugs we fix.

How current is it? A scheduled job rebuilds from JMdict, KANJIDIC2, Waller and Tatoeba every month.

A note on JLPT levels

The Japan Foundation does not publish vocabulary, kanji or grammar lists for the current (post-2010) JLPT. Word and kanji levels come from Jonathan Waller's lists, the community standard. Grammar levels follow the consensus of common JLPT preparation materials. All of them are approximations, not a guarantee of what appears on the test.

Roadmap

• Richer English meanings from JMdict alongside Waller's short glosses
• Audio (native or TTS)
• More example sentences for grammar points

Ideas and pull requests are welcome: open an issue.

Built with OpenJLPT

Made something with the data, an app, a bot, a deck or a study tool? Add it here with a pull request, or tell us in an issue.

• Your project here

Contributing and support

Found a wrong reading or an unnatural sentence? Every page on the website has a Report an error link, and CONTRIBUTING.md explains how fixes flow into the data. If OpenJLPT saved you time, ⭐ star the repo so others can find it, or sponsor the project.

License

Dataset and code: CC BY-SA 4.0. Upstream attribution (JMdict and KANJIDIC2 from EDRDG, Jonathan Waller's lists, and Tatoeba) is in NOTICE.md. To cite OpenJLPT, use GitHub's "Cite this repository" button, which reads CITATION.cff.