moji. Get the app

Updated August 4, 2026

How many Japanese words do you need to be fluent?

There is no honest single number, but there is an honest shape. In Japanese, the most frequent 1,000 words cover roughly 60% of running text, and the most frequent 5,000 still do not reach 82%. Reaching 90% coverage is said to take around 10,000 words (『図解日本語』, Zukai Nihongo, “Japanese Illustrated”, Sanseido, 2006). Meanwhile the coverage level researchers associate with comfortable unassisted reading is 98%, not 90%, because 90% coverage means one unknown word in every ten.

So the practical answer splits in two. A few thousand words is enough to make Japanese usable: to follow conversation, read graded material, and get the gist of things. The vocabulary size behind effortless native-level reading is several times larger than that. The first number is close. The second is a multi-year project, and most vocabulary advice online quietly conflates them.

What “coverage” actually means

Coverage is the percentage of the running words in a text that you already know. It counts tokens, not unique words, so common particles and verbs count every time they appear. That is why the first few hundred words are so productive and why the curve flattens so fast afterward.

The important thing is that coverage percentages sound much better than they feel. Hu and Nation tested this directly by replacing known proportions of words in a text with nonsense words. At 80% coverage, nobody reached adequate comprehension. At 90%, a small minority did. At 95%, meaning one unknown word in twenty, still only a few more. Fitting a regression to the results, they estimated that around 98% coverage, one unknown word in fifty, is needed for most learners to comprehend a text unassisted (reported in Nation, 2006).

Keep that scale in mind for every number below. 86% coverage is not “most of it.” It is roughly one word in seven that you have never seen.

Japanese has a flatter curve than English

This is the part that surprises people coming from a European language. The coverage table in 『図解日本語』 puts the top 1,000 words of English, French, and Spanish at about 80% of general writing and broadcast material. The top 1,000 Japanese words cover about 60%. At 3,000 words those European languages are at 90% or better, while Japanese has not reached 82% even at 5,000 words.

The explanation the authors give is lexical variety: Japanese frequently offers a native word, a Sino-Japanese word, and a loanword for the same thing, each with slightly different use. Their example is one object on a table being called 湯飲み (ゆのみ, teacup), 茶碗 (ちゃわん, rice bowl or teacup), and カップ (kappu, cup) by three different speakers in the same short conversation. Every one of those is a separate word you have to know, and the usage is split three ways rather than concentrated in one high-frequency item.

Practically, this means English-derived vocabulary targets undershoot for Japanese. Nation’s widely cited estimate is that 98% coverage of written English takes an 8,000 to 9,000 word-family vocabulary, and spoken English 6,000 to 7,000 (Nation, 2006). Those figures are English-specific, from the British National Corpus, and should not be transferred to Japanese directly. The Japanese equivalents are higher.

Real Japanese numbers, measured on real texts

The most useful modern dataset comes from Honda (2019), who built a 10,000-word reading vocabulary list from the Balanced Corpus of Contemporary Written Japanese and then measured what it actually covered in texts written for native readers. Her coverage figures, by vocabulary level:

Words knownNewspapersWebNovelsCasual speech
2,00055%58%62%65%
4,00070%71%73%78%
6,00077%78%79%81%
8,00083%83%83%85%
10,00086%85%86%86%

One caveat matters a lot here, and Honda states it herself: this list contains content words only, with function words excluded. Real coverage including particles, the copula, and other grammatical machinery would be meaningfully higher at every level. Read the table as the shape of the curve rather than as your personal comprehension score.

That shape is brutal at the top end. On novels, the first 2,000 words buy 62 points of coverage. The next 2,000 buy 11. The 2,000 after that buy 6, then 4, then 3. By the time you are learning your ten-thousandth word, you are paying two thousand words for about three percentage points, and you are still one word in seven short of the text.

The same study measured old JLPT reading passages, where the picture is friendlier because the texts are graded. At 2,000 words, Honda’s list covered 88% and 89% of level 4 and level 3 reading texts, but only 74% of level 1 texts. At 6,000 words it reached 96% and 97% of the beginner levels and 89% of level 1.

For the far end of the curve, Matsushita’s corpus work is the usual citation. As reported in a 2024 study in System, Matsushita (2014) put 95% coverage of texts written for native Japanese readers at about 9,500 lexemes, and 98% coverage at roughly 20,000. That upper figure is what “fluent reading of anything you pick up” actually costs, and it is worth knowing so you can decide not to aim there yet.

What each milestone feels like

Translating the curve into experience, with the numbers above as the anchor:

500 words. Not a coverage story yet. This is where sentence patterns start working: you know the particles, the handful of verbs that do most of the work, and enough nouns to make the grammar visible. You will not read anything unassisted, but you can start recognizing structure in what you hear.

1,000 words. Around 60% of running text. In practice you catch the frame of a sentence and lose the content. Slow, patient conversation about everyday topics starts to work, because a cooperative speaker will rephrase. Native material stays out of reach.

3,000 words. Comfortably inside beginner and lower-intermediate graded material, and well past the point where studying feels abstract. On native text you are somewhere in the sixties to low seventies, which is enough to follow a familiar topic with a dictionary and not enough to read for pleasure. Conversation is noticeably better than reading at this level, because spoken Japanese repeats a smaller set of words. Honda found casual speech consistently ahead of newspapers at every level, by about ten points at 2,000 words.

7,000 words. You are in the flat part of the curve, roughly the low eighties on native content words. Ordinary reading works with occasional lookups, most conversation is comfortable, and the words you are missing have shifted from “common words I should know” to “specific words for this particular topic.” This is where dictionary lookups stop breaking your reading and start merely slowing it.

Past that, the remaining gains come from reading widely rather than from a word list, because which words you need next depends entirely on what you read. That is also the point where a curated catalog stops being the right tool.

Where Moji sits on this curve

Moji - Learn Japanese Daily for iPhone ships a curated catalog of 7,435 words, every one with a hand-checked example sentence. The study order is derived from JPDB frequency data and then hand-edited, with core particles pulled forward on purpose: は (wa, topic marker) enters around position 63, が (ga, subject marker) around 89, を (o, object marker) around 129, each introduced with example sentences chosen to show where the particle sits. There is more on that in why word order matters and the most common Japanese words list.

The catalog size is a deliberate choice about this curve, not a limit we ran into. 7,435 words lands in the zone where each additional word still buys real comprehension and before the stretch where two thousand more words buy three percentage points. Getting a learner to the flat part of the curve with good example sentences and accurate meanings is a solvable problem. Chasing dictionary completeness is not, and it would mean shipping thousands of entries nobody checked. Our curation standards explain what “checked” means here.

We deliberately do not publish a coverage percentage for our own catalog. Our word list has never been measured against a Japanese corpus the way Honda measured hers, and quoting a number we did not measure would be exactly the kind of claim this page exists to push back on.

For how long all this takes in calendar time rather than word counts, see how long it takes to learn Japanese.

The honest summary

If you want a target: 3,000 words makes Japanese usable, 7,000 or so makes it comfortable for everyday material, and true unassisted reading of anything sits far beyond both. If someone quotes you a tidy figure like “1,000 words gets you 80% of Japanese,” check whether it came from English data. In Japanese, that 80% costs several times more.

If you want the frequency-ordered path with sentences that were actually checked, Moji is free to try, and the whole comprehension path stays free: flashcards, mnemonics, example sentences, kana writing, and the daily quiz. Download Moji on the App Store

Sources