VerbamorThe best language learning app, ranked by what it makes you do
I haven't used most of these apps, and I'm not going to pretend otherwise. What follows is built from each company's own documentation, read on 21 September 2026, sorted on the one axis that predicts whether you'll be able to speak: does the app hand you the answer, or make you produce it?
Someone on a 200-day Duolingo streak gets to Madrid, orders a coffee, and then stands there while the waiter asks a follow-up question. The words were all in there somewhere. None of them came out.
That gap is the thing you're actually shopping for when you type "best language learning app" into Google, and it's the thing every roundup on the first page declines to measure.
What this page is, and what it isn't
I build a language app. That's a conflict of interest, so here's how I've handled it: Verbamor sits in the same comparison as everything else, gets the same "don't buy this if" line, and I've put its weakest column in writing.
I also haven't run a six-week trial of nine apps, and nobody should believe me if I claimed I had. Everything below comes from the companies' own manuals and help centres, fetched on 21 September 2026, with links so you can check.
Which turns out to be a competitive position rather than an apology. Reviewed grades Busuu, Memrise and Lingoda from "manufacturer information and reputable third-party reviews" while filing them under a heading that says tested. I'd rather say which parts I read.
Why there's no price table
I tried to build one. It's the single most useful thing a page like this could carry, and I couldn't do it honestly.
Here's what happened on 21 September 2026. Babbel's support centre returned HTTP 403 to an automated read, and its pricing path returned 404. Duolingo, Pimsleur and Rosetta Stone all render their prices in your browser after the page loads, so the server sends no figure to read. Apple's public lookup API reports all four as free downloads and exposes no in-app subscription price at all.
So when a roundup shows a clean grid of monthly and annual figures for all of these, ask where those numbers came from. TechRadar's, updated this year, carries no dollar figure anywhere and calls apps "pricey" instead.
The one first-party subscription figure I could confirm is Pimsleur's own, in a Pimsleur blog post: "Both Pimsleur Premium and All Access subscriptions cost around $20 per month." Deliberately approximate, and it's theirs.
The other is Anki's, because Apple publishes it as an app price rather than an in-app purchase: AnkiMobile is US$24.99, one time, confirmed against Apple's own listing today.
Check the price in your own App Store, on your own account, before you subscribe. Regional pricing and promotional sales mean any figure a page prints was already wrong for somebody.
The axis nobody scores
Every app in this category does one of two things to you, and the difference decides whether you'll speak.
Recognition is picking the right answer from options on screen. Tap the tile that matches, choose from four words, arrange the jumbled sentence. The answer is present while you respond.
Recall is producing the answer with nothing to copy from. You see "the forest" and you have to find bosque in your own head before anything reveals it.
These feel similar and they are not remotely equivalent. Karpicke and Roediger gave students 40 Swahili-English pairs and split them by what happened after the first correct answer. A week later, the conditions that kept testing the pairs recalled about 80%. The conditions that kept studying them recalled 36%. Same students, same words, same time in the lab.
Now apply that to a lesson screen. If the target sentence is visible while you answer, you're in the 36% column. Duolingo's own description of its speaking exercises has learners repeat words and sentences that are shown to them, which puts the retrieval step on the app's side rather than yours.
Kang, Gollan and Pashler tested exactly this on spoken output. Learners heard Hebrew object names, and one group repeated each name after hearing it while the other tried to produce it from memory first. At a 2-day delay the retrieval group had better comprehension, better ability to produce the words, and no loss of pronunciation quality. Worth knowing the limit: that test was naming single pictured objects, so it doesn't prove anything about producing a word inside a fast sentence.
The apps, sorted by that axis
No prices, for the reason above. Every claim here is from the company's own documentation.
| App | What it mainly trains | Don't buy it if |
|---|---|---|
| Anki | Recall. You write the card, so you decide whether it asks for production. Ships SM-2 by default; FSRS is a checkbox at the bottom of deck options. | You won't actually build cards. It ships no language content whatsoever. |
| Pimsleur | Recall, spoken. Its published method names Anticipation and Graduated Interval Recall: you're prompted to produce before you hear the answer. | You need reading and writing, or you can't give it your ears for 30 minutes. |
| Babbel | Mixed. Review runs a fixed six-step ladder: 1 day, 4, 7, 14, 60, then six months, and stops. Speech recognition grades your voice. | You want to review your own words. Babbel says the review vocabulary "is predefined based on the lessons you complete." |
| Duolingo | Mostly recognition. Real published scheduling research behind it (half-life regression, ACL 2016), and most exercise types show you the answer. | Speaking is your goal and it's your only tool. Its strongest published result is a reading and listening one. |
| Memrise | Recognition-led, on a fixed ladder: 4h, 12h, 24h, 6d, 12d, 48d, 96d, 6 months, identical for every user and every word. | You want the schedule to respond to how you personally do on a given word. It can't. |
| Quizlet | Recognition by default. Its Progress tracker buckets terms by how often you answered right, with no due dates and no intervals. | You want the app to tell you what to review today. It never computes a next-review date. |
| Verbamor | Recall. Cloze cards from your own lessons, FSRS scheduling, answer hidden until you commit. | You're a beginner. It holds words you've already met and won't teach you your first thousand. |
Two honest notes on that table. Babbel wins an axis outright that I can't match: it states that all 100,000 hours of its content are voiced by native speakers, where Verbamor's card audio is generated by text to speech. And Duolingo's audio is also text to speech, built from recorded actors, so that's a tie between us rather than a point for me.
The streak trap
A streak counts days you opened an app. That's a number the app benefits from regardless of whether you learned anything, which is worth sitting with before you let it stand in for progress.
The mechanism isn't sinister, just misaligned. Daily engagement really is correlated with learning, so the metric isn't fake. It's that an app optimising for it will always find it cheaper to make the session pleasant than to make it hard in the way that builds memory.
And the hard way is the one that works. Elizabeth and Robert Bjork's term for it is desirable difficulties, and their list names "using tests, rather than presentations, as study events" directly. Their caveat travels with it: if a learner lacks the background to respond successfully, the difficulty stops being desirable.
So here's what to check before subscribing to anything. Open the app and answer three cards. Was the correct answer visible on screen while you responded? If yes, you've bought recognition practice, whatever the marketing said. That's a real product and it will not, on its own, get you through the follow-up question in Madrid.
Which one for your language
Coverage settles this before quality gets a vote, and it's where the big roundups are least useful to anyone outside the obvious five languages.
Babbel publishes 14 languages, and a subscription covers all of them rather than one. Pimsleur publishes 50-plus courses, which makes it the realistic option for less commonly taught languages. Anki covers everything and nothing: no content ships, so it's only unlimited if you're willing to build.
Verbamor covers five: Spanish, French, Italian, Portuguese and Russian. If you're learning Japanese or Korean, that's the end of the conversation for my app, and Anki or Pimsleur is where I'd send you.
One caution on the "learn any language" claim generally. The State Department's own current figures put Category I languages, the ones closest to English, at 552 to 690 class hours over 24 to 30 weeks, on a week of 23 class hours plus 17 hours of self-study. Spanish sits at 30 weeks. No app's daily session is denting that on its own, and the numbers most blogs still quote for this are the retired ones.
Some of you should go install Anki
If you'll genuinely build your own cards, stop reading and install Anki. It's the correct answer to this entire page for a certain kind of person, and I'd give a friend that answer without hesitating.
Free on Windows, macOS, Linux and Android, free syncing through AnkiWeb, and iOS is the only platform that costs anything at US$24.99 once. Turn FSRS on at the bottom of the deck options page, because the manual calls it "an alternative to Anki's legacy SuperMemo 2 (SM-2) algorithm" rather than the default. Plenty of articles get that backwards.
Know the real cost before you commit. Anki's own manual warns that "if you are consistently learning 20 new cards a day, you can expect your daily reviews to be roughly about 200 cards/day," and that more than one user "has excitedly studied hundreds of new cards over their first few days" and then drowned. The sticker price is the cheapest thing about it.
Verbamor exists for the narrower case where card-making is the bottleneck. You record or type what came up in your tutor lesson and it builds the cloze cards from that, on FSRS, in the five languages above. It's $19.99 a month, or $129.99 a year with a 7-day trial on the annual plan, and there's no free tier, so if you want to try before paying, that trial is the only door. Both are US prices, read off the app's own plan definitions on 21 September 2026; the App Store sets a different number in other countries. It teaches you nothing new, which is the point and also the limitation: it holds what your tutor already taught you. The case for doing it in Anki instead is a real one and I've written it out.
Questions people actually ask
What is the best language learning app?
There isn't one, and the question is doing you harm by sounding answerable. Sort the category by what the app makes you do instead. Apps that ask you to tap the right tile from four options train recognition. Apps that make you produce a word from a bare meaning prompt, with nothing on screen to copy, train recall. Recall is the thing that shows up in conversation. Pick on that axis first, then on whether the app covers your language, then on price.
Why won't this page print a price comparison table?
Because I could not verify the prices. On 21 September 2026 I tried to read them first-hand: Babbel's support centre returned HTTP 403 and its pricing path returned 404, and the Duolingo, Pimsleur and Rosetta Stone pricing pages all render their figures in the browser after the page loads, so the server sends no number to read. Apple's public lookup API reports every one of those apps as a free download and exposes no in-app subscription price. Any roundup printing a tidy table of these figures copied them from another roundup. Check the price in your own App Store before you subscribe.
Does a daily streak mean an app is working?
No. A streak measures how many days you opened the app, which is a measure the app benefits from whether or not you learned anything. The useful question is what happens inside the session. If the answer is visible on screen while you respond, you are practising recognition and the streak is counting reps of the easier task. Karpicke and Roediger found that repeated studying after the first correct answer had no effect on recall a week later, while repeated testing produced a large positive effect.
Which app is best for my specific language?
Coverage decides this before quality does. Babbel publishes 14 languages, so a subscription covers all of them but the list is short. Pimsleur publishes 50-plus courses. Anki ships no language content at all, which makes it unlimited and also means you supply everything yourself. Check the published course list for your target language first, because an excellent app that doesn't teach Russian is not an option for someone learning Russian.
Should I just use Anki?
If you will genuinely build your own cards, yes, and it is the honest answer to this whole page. Anki is free on Windows, macOS, Linux and Android, syncing through AnkiWeb is free, and iOS costs US$24.99 once rather than monthly. It ships no language content, so you are buying an engine and supplying the Spanish yourself. Turn FSRS on at the bottom of the deck options page, because the manual describes it as an alternative to the legacy SM-2 default rather than the default itself.
Sources
Recognition versus recall
- Karpicke, J. D., & Roediger, H. L. (2008). The critical importance of retrieval for learning. Science, 319(5865), 966–968. The 80% and 36% figures are from the paper's results; 40 Swahili-English pairs, final test one week later.
- Kang, S. H. K., Gollan, T. H., & Pashler, H. (2013). Don't just repeat after me: Retrieval practice is better than imitation for foreign vocabulary learning. Psychonomic Bulletin & Review, 20(6), 1259–1265. Hebrew nouns, picture-naming at a 2-day delay.
- Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In Psychology and the Real World, 56–64. Source of the definition and of the caveat that difficulties stop being desirable without the background to meet them.
What each app documents about itself
- Anki Manual. Deck options. Read 21 Sep 2026. Source of the FSRS-is-an-alternative quote and of the 20-new-cards / 200-reviews warning.
- Pimsleur. Pimsleur cost breakdown. Read 21 Sep 2026. The only first-party subscription figure on this page: "around $20 per month."
- Apple. AnkiMobile Flashcards listing, US$24.99, confirmed via Apple's public lookup API on 21 Sep 2026.
- Settles, B., & Meeder, B. (2016). A trainable spaced repetition model for language learning. ACL 2016. Duolingo's published half-life regression. Whether it still runs in production is not something I can verify from outside.
- Babbel's six-step review ladder, its statement that review vocabulary is predefined by lessons completed, its 14 languages and its native-speaker audio claim are from Babbel's own support centre and about page. Its pricing pages did not respond to an automated read on 21 Sep 2026.
- Memrise's fixed interval ladder and Quizlet's Progress tracker are documented in their respective help centres. Neither publishes a price I could confirm first-hand.
How long it actually takes
- U.S. Department of State, Foreign Service Institute. Foreign language training. Category I: 24-30 weeks, 552-690 class hours, on 23 class hours plus 17 hours of self-study a week. The widely quoted 600-750 figure is from the retired page.
Every load-bearing claim Verbamor makes is traced to its paper on the research page.
Cards that make you produce the word.
Verbamor builds cloze cards from your own tutor lessons and schedules them with FSRS, in Spanish, French, Italian, Portuguese and Russian.
Start 7 days free