"How many words do I need?" has a tidy answer in a language with one script and one reading per word. Japanese does not work that way, and counting kanji the way you would count French or Spanish vocabulary undercounts the real job by a wide margin.
One word, several readings
A word in most European languages is spelled one way and said one way. A Japanese word written in kanji usually is not. 生 alone can be read い (i) in 生きる (ikiru, to live), なま (nama) in 生ビール (nama bīru, draft beer), or せい (sei) in 先生 (sensei, teacher) — three different sounds from the same character, depending entirely on which word it is standing in for that day. Learning "the word for teacher" and learning "the character 生" are two different tasks that happen to overlap, and a vocabulary count that treats them as one number is hiding the actual work.
Compounds do the multiplying
Most everyday Japanese nouns are two-kanji compounds built from readings you may already half-know from other words — 天 (heaven) and 気 (spirit) combine into 天気 (tenki, weather), each character keeping the on'yomi reading it uses in compounds rather than the native reading it would use alone. Once you can read the pieces, a lot of new compounds become guessable rather than new. Until then, each one looks like an unrelated word to memorise, which is why vocabulary can feel like it is growing fastest right when you have started to feel confident.
Counters attach to the numbers you already know
Counting objects in Japanese needs a counter word chosen by what kind of thing you are counting — flat things, long thin things, small animals, people and others each take a different suffix on the number. The numbers themselves are a short list to learn; the counters riding on top of them are effectively a second, smaller vocabulary you need before "three" is useful for anything more specific than counting on your fingers.
Recognising a word and producing it are further apart here
In many European languages, if you can read a word and understand it spoken, you are close to being able to say it. Japanese widens that gap. Pitch accent — a rise and fall in pitch across a word, rather than stress on a syllable — can be the only thing separating two words that look identical in hiragana: 橋 (hashi, bridge) and 箸 (hashi, chopsticks) are told apart in speech by pitch alone, something a page of romaji never shows you. And a word you read comfortably in kanji can still stall you mid-conversation, because recalling the sound on demand is a different skill from recognising the shape on a page.
Put those together — multiple readings per character, compounds that hide or reveal their pieces depending on what you already know, counters as a second small vocabulary, and pitch accent splitting recognition from production — and a single number was never going to describe the task honestly. Our own word database, the same one behind every language in the app, still sorts Japanese vocabulary into CEFR bands like any other language: 685 words at A1 up to 654 at C2, out of 9,413 keys across 665 themed packs. The bands are a reasonable order to learn in. They are just a much rougher map of the distance than they would be for a language with one script and one reading per word.
Watch Your Language does not ask you to hit a target number. It shows a few Japanese words at a time, in their actual script with a romaji reading alongside — 天気 (tenki) — on your iPhone Lock Screen and Apple Watch face, and brings back the ones you are still shaky on more often than the ones you already have.
A more useful question
Not "how many words", but "how many readings of the words I already have". Every time a kanji you know turns up in a new compound, that is progress on the same character, not a new item starting from zero. Track it that way and Japanese vocabulary looks less like an endless list and more like a smaller set of pieces recombining — slowly, then quickly.
