More than ten million people currently hold a Duolingo streak of at least 365 days. That number was made public in early 2026, and it is worth sitting with for a moment, because in consumer software it is close to absurd. Getting five percent of your daily users to come back seven days running is considered a triumph. Ten million people, a full year, no days off.

Whatever you think about owls, that is an enormous achievement. It is also the reason this comparison is more interesting than the usual app fight. Duolingo did not win by being the best teacher. It won by solving a problem the teaching profession has never solved: how to make somebody show up tomorrow.

Vokabulo is built on a different bet. Not a better one in the abstract, a different one, aimed at a different person at a different point in their learning. This piece is about where those two bets actually diverge, and why it is good news for learners that both exist.

A Commodore 64 in Guatemala City, and an English test that cost 1,200 dollars

Luis von Ahn was born in Guatemala City in 1978 and raised by his mother, one of the first women in the country to finish medical school. When he was eight she bought him a Commodore 64. He has described the English-language school he attended as a great privilege, and the word privilege is the hinge of the whole story: in Guatemala, learning English was something you bought.

The detail that sticks is what happened when he applied to universities in the United States. To sit the TOEFL exam he had to travel to El Salvador, and the trip cost him more than 1,200 dollars. An English test, a plane ticket, and a bill larger than many families' monthly income, all to prove he could speak a language he already spoke. He later described the testing industry as extractive and ripe for disruption, and he was not wrong.

He went to Duke, then to Carnegie Mellon for a PhD under Manuel Blum, writing a dissertation on what he called human computation. Along the way he invented CAPTCHA, and then reCAPTCHA, the version that made you squint at a smudged word from a scanned book. That second one was quietly brilliant: while proving you were human, you were also digitising books nobody could afford to digitise by hand. Google bought it in 2009. He had a MacArthur Fellowship by 2006, at 27.

His co-founder was Severin Hacker, a Swiss graduate student in his lab, who has spent fifteen years repeating the same sentence in interviews: education should be free. Two people, one from a country where English was a paid product and one from a country where it mostly was not, agreeing on the same thing from opposite directions.

The anecdote worth telling: Duolingo was going to translate the internet

Here is the part most people have forgotten. Duolingo's original business model had nothing to do with subscriptions.

The plan was human computation again, applied to language. Learners would practise by translating real sentences from real documents, and the customers paying for those translations would keep the app free forever. It was not a thought experiment: CNN and BuzzFeed both paid Duolingo to translate their articles, and the exercises learners did were the work product. You practised your Spanish, and somewhere a news article quietly became available in another language.

It was elegant, it was true to the reCAPTCHA idea, and it did not survive. Learner translations were uneven, the volume was hard to predict, and the whole thing was eventually retired in favour of subscriptions, ads and the Duolingo English Test, which is itself a direct descendant of that 1,200 dollar trip to El Salvador.

I like this story because of what it says about the company. They kept the mission and threw away the mechanism. Most startups do the reverse.

What Duolingo perfected

The streak is the most copied piece of design in consumer software, and Duolingo did not invent the idea so much as finish it.

The finished version is a system, not a number. There is the streak itself, the freeze that saves it when life happens, the friend streaks that add mild social debt, the notifications that became a genuine internet meme, and a mascot the company was willing to kill off in a publicity stunt in February 2025 and resurrect once users had earned enough experience points to bring him back. In 2025 they went further and rebuilt the company around AI, which produced real backlash, including people deleting the very streaks the whole machine exists to protect.

And it keeps working outside language entirely. Duolingo now teaches music, maths and chess, and by early 2026 the chess course alone had around seven million daily users. That tells you what the core asset really is. It is not a Spanish curriculum. It is a habit engine that happens to have Spanish loaded into it.

The company's numbers follow from that. Revenue crossed a billion dollars for the first time in 2025, landing at 1.04 billion, up 39 percent on the year. When Duolingo reported the second quarter of 2026 in August, it posted 298 million dollars of quarterly revenue, 58.7 million daily active users, up 23 percent year over year, 140.6 million monthly actives and 12.7 million paying subscribers. This is the most successful edtech company ever built, and it got there honestly, by being extremely good at one very hard thing.

The awkward middle: what the evidence actually shows

Now the part that is usually written unfairly, so let me try not to.

The research on app-based language learning is genuinely mixed. A 2017 study comparing schoolchildren learning Spanish through Duolingo's gamified approach against a traditional classroom found no significant difference between them. A 2022 study found that people who completed Duolingo courses had reading and listening proficiency comparable to university students after four semesters, which is a striking result, and also one produced with the company's own involvement, as most efficacy research in this industry is.

Read together, the honest summary is something like: it demonstrably does something, the something is strongest in reading and listening, and the evidence for speaking the way a person actually speaks is thinner everywhere, for everyone, including us.

The reason is not corporate cynicism. It is that engagement is measurable and proficiency is not. A company can see your streak, your lesson completions and your retention curve on a dashboard, today, per user. Whether you held your own at a dinner in Madrid last Thursday is invisible to every analytics system ever built. So the entire category, ours included, optimises what it can see. Any product that runs on daily active users will drift toward the metric it can count, and the interesting question is not whether that pull exists but what you do about it.

What I would resist is the fashionable conclusion that gamification is therefore a trick. Encouragement is not the problem. Earned encouragement is one of the most useful things software can give a learner, and anybody who has tried to study alone for six months knows it. The thing worth being careful about is applause for merely opening the app, which is a different thing wearing the same clothes.

Where the two products actually part company

Here is the sentence the whole comparison rests on: the first thousand words belong to everybody, and the next five thousand are yours.

At A1 and A2 this is not a limitation, it is a gift. Bread, water, train, tomorrow, tired, expensive, on the left. Every beginner in every language needs almost exactly the same list, in almost the same order, and a single curriculum can serve millions of people identically without cheating any of them. At that stage the bottleneck is not relevance. The bottleneck is turning up, and Duolingo owns turning up.

Then something changes, usually around B1, and it changes quietly enough that most people misread it as a plateau in their ability rather than a shift in the problem.

Picture two German speakers who both learned French. One is 23 and doing a year in Lyon: her French runs on coloc, partiels, rendre un devoir en retard. The other is 55 and sits in meetings in Paris: his French runs on levée de fonds, conseil d'administration, à titre indicatif. Their beginner vocabulary was identical. Their advanced vocabulary has almost nothing in common. No curriculum, however well designed, can serve them both, because the thing they now need is not more language in general. It is their language in particular.

There is a second shift on top of the first. Early on you are missing words. Later you have the words and they are the wrong ones. You say happy when you mean relieved, or delighted, or quietly pleased. You say not good when you mean not ideal but acceptable. Nobody corrects you, because you were understood, and being understood is exactly what stops anyone from telling you that you sounded blunter, or vaguer, or younger than you are.

Learners describe this better than marketers do. The line I keep coming back to is: I am not a beginner, I am just tired of sounding simpler than I am.

Two products, two shapes

Once you see the problem that way, the design differences stop looking like competitive positioning and start looking like arithmetic.

A course app has to decide what everybody learns next, so it needs a curriculum, and a curriculum needs an order, and an order needs to be the same for everyone. That is coherent, and it is also why the useful part of a course thins out as you climb. There is no sensible linear syllabus for C1, because at C1 there is no shared next thing.

Vokabulo starts from the other end. The words come from your life, not from a list: the phrase your colleague used in a meeting, the sign you walk past every morning, the word in the rental contract you half understood. You add it in ten seconds, with the context you met it in, and spaced repetition brings it back on a schedule instead of a streak. Words carry usage labels, so you know whether a phrase belongs in a message to a friend or in an email to a client, which is the difference between knowing a word and being able to use it without flinching.

That design has an obvious cost, and it is worth stating plainly: it needs you to already have a language to be precise in. Vokabulo is not where you learn that the French for bread is pain. It is where you work out why je suis content and je suis ravi are not the same offer of thanks. At A1 that distinction is noise. At B1 it is the whole job.

So the categories are not really competing. Courses teach you the language. Native content exposes you to it. Flashcards help you remember it. What none of them quite answers is the request underneath all of it, which is help me become an articulate speaker. That gap is where we work.

It is genuinely good that you have a choice

If you are at zero in a language, go and use Duolingo. I mean that without any hedging. You need a habit before you need a strategy, the app is free, and it is the best habit builder anyone has made. Buying a precision tool before you own a hammer is a way of not starting.

If you have been at it a few years, if you get by, if people have stopped switching to English with you but you still hear yourself reaching for the same forty adjectives, then the thing holding you back is not discipline. You have plenty of discipline, you have a 400 day streak. It is that the words you need next are not on anybody's list, because they are yours.

The healthiest thing about this market in 2026 is that it no longer has one shape. There is an app for starting, and there are apps for what comes after starting, and nobody has to pretend that one product should carry a learner from bonjour to negotiating a contract. Duolingo built the on-ramp better than anyone, at a scale that puts language learning in front of hundreds of millions of people who would never have walked into a classroom. That is not a rival position to defend against. It is the reason there are so many people at B1 to build for.

If you want the longer version of the vocabulary argument, we wrote it up in why Duolingo doesn't stick and in a broader comparison of the alternatives, and the shift that happens around B2 has its own piece on breaking the intermediate plateau.