Free JLPT Anki Deck: 10,600 Japanese Vocabulary Cards N1–N5 with Audio

Some time ago I was working on a simple quiz app to learn some Japanese vocabulary. I asked my agent to create a list of words and it did, but then I noticed some of the readings were incorrect. Not obviously wrong, but just slightly different, so you’d learn wrong pronunciation without even realizing it. So I went on a quest for some reliable data.

This JLPT deck by “egg rolls” contains all JLPT levels (N5–N1) with ~10,600 words. All words have real voices (not synthetic), at least one sentence with audio, pitch accents, POS tags. Words are taken from real JLPT exams from 2010 to late 2025, so for higher level words you get actual exam frequency. That’s real data, not made up.

Before downloading, please note that the deck UI and glosses are in Chinese (simp./trad.), but the actual Japanese vocabulary does not depend on your native language. Words, readings, sentences, voices — everything is in Japanese. But if you import this into Anki and expect English definitions, you’ll be disappointed. If you want to use this as a source of structured data for your app, your agent can easily provide English/Russian translations on top of the Japanese ones.

1How to install the JLPT deck in Anki

After importing, you’ll have structured sub-decks for each JLPT level (N5–N1). Front side shows the word, back side has reading, meaning (in Chinese), sentence with audio and similar/opposite words. You can sync this deck with AnkiWeb for multi-device sync just like any other Anki deck.

2How to use the vocabulary deck as data for a quiz app

If you’re developing a Japanese quiz app or learning tool, the easiest way is to provide your agent with a link to the GitHub repo and have it scrape the data. This deck is available as a .apkg file (which is basically a zipped SQLite database), so your agent can unpack the words and work from there. Alternatively, you can send your agent to ankiweb.net/shared/info/832276382 to see what words are included and plan your app accordingly before writing any code. In both cases, you’ll get a reliable data source instead of some made up stuff your agent might come up with.