JPDB vs immit: Which Japanese Reader Fits Your Workflow?
You found JPDB, you pasted in a chapter of your current novel, and within seconds it handed you a deck of every word you don't know yet, ordered by how often it shows up. That is a genuinely good moment. Then you tried to read the actual novel on a different site, hovered a word you'd forgotten, and nothing happened, because JPDB lives on jpdb.io and the page you're reading does not.
That gap is the whole comparison. JPDB and immit both want to help intermediate-to-advanced learners read more Japanese and remember what they read, but they take opposite routes. JPDB is a deck-and-database platform you go to. immit is a popup dictionary that comes with you onto whatever page you're already reading. This post lays out which one fits which workflow, where they overlap, and why a lot of learners end up using both.
JPDB (jpdb.io) is a free, all-in-one web application for Japanese learners that combines a frequency database, prebuilt vocabulary decks, and a machine-learning spaced repetition algorithm. You paste in any text, or pick from its library, and it will extract vocabulary automatically, rank each new word by frequency, and schedule it for review.
The standout asset is the library. As of June 2026, JPDB lists 21,434 prebuilt decks drawn from 1,399 anime, plus novels and other media, with over 130 million example sentences behind them. You can sort a deck by global frequency (how common a word is across all Japanese) or by local frequency (how common it is inside the specific show or book you're about to watch), so you can pre-learn the vocabulary for an episode of your favorite anime before you press play. JPDB also assesses the difficulty of media using language models and suggests what to read or watch next based on how many words you already know.
It is built for intermediate-to-advanced learners. The whole model assumes you already have grammar and kana and want to scale vocabulary fast against the real native material you consume, so the practice loop is reading or watching first and reviewing in context second.
The frequency-first, deck-based approach is its real strength, and it's worth naming clearly before getting into where immit differs.
JPDB's parsing is the draw. It can extract vocabulary from any text you paste, break each result down into base kanji and visual components, and tell you which words are new versus previously learned across every deck you've touched. Because it tracks your known-word database globally, a word you learned from one anime is marked as known when it shows up in a novel, so you're never re-grinding the same common words. Its vocabulary cards and kanji cards stay linked to that single known-word count.
The difficulty and comprehension metrics are also unusually good. JPDB analyzes your known-word database to estimate how much of a given show or book you'd understand, which turns "what should I read next" from a guess into a number. For data-oriented immersion learners, that feedback loop is the appeal, and immit does not try to replicate it.
JPDB's spaced repetition is a modern machine-learning algorithm rather than the SM-2 scheme most older flashcard programs inherited from the 1980s. JPDB owns the "smart algorithm" narrative in this niche, and we won't pretend otherwise.
These are trade-offs, not failures. JPDB made deliberate choices to be a platform, and the costs of those choices are exactly where a different tool fits.
JPDB has no popup dictionary for the open web. Its browser extension and public API parse and track text, but the core reading-and-saving experience happens on jpdb.io. When you're reading a news article or a blog and you hit an unknown word, you're tab-switching or copy-pasting to look it up. JPDB also requires a login to save progress and run reviews, runs web-only with no offline mode, and has no native mobile app, so reading on a plane or a phone without a connection isn't part of the model. The interface is functional and dense rather than modern, which long-time users stop noticing and new users notice immediately.
None of that makes JPDB worse. It makes it a different shape. The extension does work properly for what it's built to do, parsing and tracking text, but the moment a Japanese word shows up on a website outside JPDB's own ecosystem, the lookup-on-the-spot experience isn't there. If your reading happens inside JPDB, the gaps barely register. If it happens out on the open web, they're the whole experience.
immit is a popup Japanese-English dictionary with a built-in 8-stage spaced repetition system, available as a Chrome extension and as a desktop app for Mac, Windows, and Linux. You install both, the extension for in-browser reading and the desktop app for everything else. Hover a word on any web page, see the reading, part of speech, definition, an example sentence, and pronunciation audio in about 0.1 seconds, then save it to your personal deck with one click. You can click the speaker icon to hear the word, which is the point where lookup and memory start happening in the same place instead of across two tools. There's no account required for the free tier, and lookup, saving, and review all work offline. Pro is $9/month, $108/year, or a $299 one-time purchase, and adds multi-device sync, cloud backup, dark mode, and priority support.
The structural difference from JPDB is location. immit's popup runs on the page in front of you, including online manga readers that render text as HTML and YouTube captions, so there's no paste step and no destination site. It also ships a second surface JPDB has no equivalent for: the Pocket Dictionary, a pinned dictionary window that stays anchored in the corner of the page while you read. You can type a word into it to look something up actively without leaving the page, and you can run flashcard reviews inside it during a reading break. It's a dictionary you read alongside, not one you go to.
Where immit is honest about its limits: it does not have JPDB's prebuilt decks, and it doesn't build one for you. immit's model is mine-while-you-read. You save each new word as you actually hit it, and the deck grows from your real reading rather than from a pre-parsed list. immit also shows kanji at the word level, the headword with its reading and meaning, not as a separate per-character breakdown, and it does not display pitch accent, which is something Yomitan can do once you set it up. If pre-curated, frequency-ranked decks or character-level kanji study are what you came for, those are JPDB's strengths and we won't claim them.
| | JPDB | immit |
|---|---|---|
| Core model | Deck + frequency database you go to | Popup dictionary that comes to the page |
| Popup lookup on any web page | No (parses on jpdb.io / extension) | Yes, ~0.1s hover lookup |
| Prebuilt decks | 21,434 decks, 1,399 anime | None (mine-while-you-read) |
| Spaced repetition | ML-based algorithm | Built-in 8-stage SRS |
| Account required | Yes (login) | No, for the free tier |
| Offline | No (web-only) | Yes (lookup, save, review) |
| Native / desktop app | No | Mac, Windows, Linux desktop app |
| Price | Free, all features | Free tier; Pro $9/mo, $108/yr, or $299 once |
Verify JPDB's current feature set on jpdb.io before relying on this table; both products ship changes.
For traditional sentence mining, where you build i+1 cards from sentences you encounter, Anki is still the most customizable destination, and JPDB can partially import an existing Anki deck so you don't start from zero. Anki gives you full control over your anki cards, offline access, and it's open source and free except for the iOS app. JPDB gives you automatic vocabulary extraction and frequency ordering, so it front-loads the deck-building that Anki makes you do by hand.
immit sits to the side of both. It isn't a deck-import tool and it isn't trying to be Anki. It's the lookup-and-save layer for when the words come from reading on the open web rather than from a pasted block of text. If you want the full sentence-mining-into-Anki pipeline, our 5-step sentence mining workflow walks through it. If you mostly want to read and save vocab without maintaining a deck system, that's immit's lane.
Most immersion learners don't use one tool, they use a stack: a grammar resource, a popup dictionary, and an SRS layer. JPDB and immit slot into that stack differently, and it helps to see them next to the tools you may already run.
Yomitan is a Chrome extension for quick word lookups that many learners pair with Anki via an add-on. It deinflects out of the box, and once you install a pitch accent dictionary it can display pitch accent too, something immit does not do. On a Mac, the built-in Apple Dictionary allows instant word lookup but has no Japanese learning layer at all. JPDB replaces the deck-and-frequency part of that stack; immit replaces the popup-plus-SRS part in one tool, so you're not wiring a lookup extension to a separate flashcard app.
A grammar resource still sits upstream of all of them, and none of these tools cover output like writing or speaking, so plan for that separately. If you want ongoing tips on tightening this stack, immit's blog tracks how the workflow shifts as the tools change, which is worth a bookmark for the future.
Yes, and a lot of learners reasonably do. They solve different halves of the same day.
A common split: use JPDB to pre-learn vocabulary for a specific anime or novel and to track your global known-word count, then use immit for everything you read out on the web, where JPDB's popup doesn't reach. JPDB can partially import from Anki, so if you've exported decks before, that path stays open. This isn't a grudging "they both exist" coexistence. The frequency-deck workflow and the read-anywhere-popup workflow genuinely don't overlap, so running both is closer to using a map and a flashlight than to using two maps.
If you're choosing only one, the question is where your reading happens. Inside a curated, frequency-ranked deck system: JPDB. Out on whatever page you land on, with retention built in and no account: immit. You can add immit to Chrome free and test that against your own week of reading.
JPDB is a free, all-in-one web application for Japanese learners that extracts vocabulary from any text you paste, ranks each word by frequency, and schedules it with a machine-learning spaced repetition algorithm. It offers 21,434 prebuilt decks from 1,399 anime plus other media, tracks your known words across every deck, and estimates how hard a given show or book will be based on what you already know.
Yes. As of June 2026, JPDB is free with all features available, including the full deck library and the SRS. It requires creating an account and logging in to save review history and track your progress. Verify current terms on jpdb.io before relying on this.
Not in the way a hover dictionary does. JPDB's browser extension parses and tracks text, but the core reading-and-review experience lives on jpdb.io, so looking up a word on an unrelated web page generally means copy-pasting or tab-switching. A popup that runs directly on any page, the way Yomitan or immit work, is not JPDB's model.
JPDB works in the other direction more cleanly: it can partially import an existing Anki database so you don't rebuild your decks. Anki-export support from JPDB is more limited, so check JPDB's current documentation if a clean export to Anki is a hard requirement for you.
immit is the closest fit if hover lookup on the open web is what JPDB is missing for you. It's a popup dictionary plus built-in 8-stage SRS that works on any web page, including online manga readers that render text as HTML and YouTube captions, with no account and full offline support. It does not include JPDB's prebuilt frequency decks, so it's an alternative for the read-anywhere half of the workflow, not for the curated-deck half.
No. JPDB is web-only with no offline mode and no native mobile app, so it needs a connection and works best on a full browser. If offline review on a phone or a plane matters, a tool with local data and a native app, such as immit, fits that gap.
For many immersion learners, yes. JPDB handles pre-learning frequency vocabulary for specific media and tracking your global known-word count; immit handles lookup and retention for everything you read out on the web. The two workflows don't overlap, so running both is reasonable rather than redundant.
