Home / Transcripts / Duolingo, Inc. (DUOL) · November 3, 2026

Duolingo, Inc. (DUOL) Earnings Call Transcript

November 3, 2026

NASDAQ US Consumer Discretionary Diversified Consumer Services earnings 56 min

Earnings Call Speaker Segments

Deborah Belevan executive
#1

Hello everyone, and thanks for joining Duolingo's third-quarter earnings webcast this evening. We posted the shareholder letter for this quarter earlier today once the market had closed, and you can find a copy on our website at investor.duolingo.com. With us on today's call are our Co-Founder and CEO, Luis von Ahn, and our CFO, Gillian Munson. They'll begin with prepared remarks, and afterward we'll open the line for questions. Analysts who wish to ask something can do so using the raise hand feature. Keep in mind that this call is being recorded <<operator_script>>. Before we get started, please be aware that we'll be making forward-looking statements about future events and financial performance. Those forward-looking remarks involve dangers and unknowns spelled out in our SEC filings, and they depend on assumptions we consider reasonable as of today. We assume no duty to revise them. We'll also cover financial measures under both GAAP and non-GAAP. You can find reconciliations of the two in our earnings materials, and we urge you to look at them when assessing our performance. With that, I'll turn things over to Luis.

Luis von Ahn Arellano executive
#2

Thanks, Debbie, and thanks to everyone for being here. Q3 was excellent for us. DAUs sped up for a second straight quarter, reaching 29% growth year over year, compared with 23% in Q2. Bookings, revenue and profitability likewise came in above expectations. Early in the year we set out a plan putting better teaching and user growth first, aware that results would take time to appear. That user acceleration we were hoping for is now showing up, and I'm proud of the way our team has carried out our strategy; we expect this momentum to carry on, with Q4 DAU growth year over year similar to Q3 and Q4 bookings a little ahead of prior guidance. Each new learner widens the audience we can monetize down the road. Though better teaching and user growth stay our top priorities for this year, we're steadily leaning further into monetization in ways that keep the quality of our free experience intact. Our ambition is to eventually reach 1 billion learners and to build a far larger business over time. The more people we can teach, the greater that opportunity gets; with that, I'll hand it over to Gillian.

Gilian Munson executive
#3

Thanks, Luis. It's great to have everyone with us this evening. We entered the year with a bold plan to get our growth accelerating again, and the results from Q3 demonstrate that we're delivering on it. The quarter's figures are fairly clear-cut, so I'll focus in on just two items. First, let's look at how widespread our DAU and bookings growth was. Compared with a year earlier, DAU growth picked up in 8 of our 10 biggest countries. On top of that, subscription bookings growth versus the prior year reached double digits across most of the regions we operate in. Second, Q3 operating expenses contain an $11 million bonus accrual. That bonus is a reward for our team's work in getting user growth accelerating again. We're on course to meet the targets, so we've started recording the expense. A second portion of the bonus sits within our Q4 guidance. and we expect the full bonus to be paid in early 2027. Now to Q4. We anticipate bookings growth of 13% year-over-year, revenue growth of 14% and an adjusted EBITDA margin of 26.1%. To assist you in building your models, I'd like to flag just a couple of items for you to consider; we expect gross margin to decline somewhat from Q3 as Video Call is rolled out to all Paid Subscribers. Operating expenses are likely to rise more quickly than revenue, reflecting our ongoing investments and the second portion of the bonus I just referenced. Consequently, we expect adjusted EBITDA to come in roughly flat versus a year earlier at $84 million. For the full year, we expect to deliver bookings growth of 12% year-over-year and revenue growth of 18%, together with a 26.5% adjusted EBITDA margin; we intend to give our complete outlook for 2027 in February. The momentum we're seeing in our business encourages us, and we look forward to taking your questions. Over to you, operator.

Ygal Arounian analyst
#4

I suppose I'd like to begin with the growth you're seeing, which we expect will carry on through 4Q. There are quite a few levers you're pulling here. Could you perhaps quantify or qualify what the main drivers are among the new partnerships, the games, Junior's Toy Hunt, and those engagement features you're pushing? Then, as a second part to that, are there any differences in monetization, retention, or other KPIs coming from these partnerships? And should we expect more of them from you?

Luis von Ahn Arellano executive
#5

Okay. Yes. Thanks for asking. So to begin with, most of our growth this year is driven by product improvements. In the past, we've talked about our idea of The Green Machine — essentially an optimization system we have that simply keeps making the app better as time goes on. Basically, each week the app runs hundreds of new A/B/B tests that improve it a little bit. And the vast majority of our gain in active users stems from retention improvements driven by that. The product keeps getting better; now, over the last few months, we've had a few events that matter a great deal and are worth highlighting. One was the Streak Revival last quarter. This quarter, we had the partnerships with Brawl Stars and also Eggy Party. And they did bring users, but I really want to stress that we simply believe most of it comes from improvements to the app, which is why we think this will keep continuing going forward. As for the uses we get from these partnerships, they retain very well; and part of the reason is that we did some product work to make it so. So how these partnerships work is that you're a player of Brawl Stars or whatever game is partnering with us. You come to Duolingo to get rewards for that game, for Brawl Stars for example. But to get the rewards, we get you to stay multiple days on Duolingo. And what that does is it builds a habit. And so we see that these players coming in have very good retention.

Gilian Munson executive
#6

And when you look at both DAU and bookings, what we've been discussing is just how broad-based the growth was this quarter — geographically as well, and in terms of how the product is performing. We feel really good about how widespread the growth is that we're seeing in the business, across both DAU and monetization.

Luis von Ahn Arellano executive
#7

Yeah. What I mean is, growth is happening pretty much everywhere for us, and in 8 of our 10 biggest markets, DAU growth actually sped up. So that's something we're quite pleased about.

Wyatt Swanson analyst
#8

Can you give us some detail on Paid Subscriber additions compared with daily active users? Paid subs growth has kept decelerating year-over-year even though DAUs have picked back up. Is that mostly because of extended free trials? Or could you share any further color on that, and do you expect those trends to shift at all?

Gilian Munson executive
#9

I'll take the subs numbers, and then maybe you can speak a bit about the trends there. Wyatt, the way to think about the subs numbers is that they largely track revenue. So they're sort of an average in there. So subs numbers tend to track that revenue growth; so on the subs growth side, it probably bottoms in a couple of quarters as we work through the bookings from before this quarter. So as we rebuild bookings growth, that revenue growth will follow, and those subs growth numbers will follow too. So it's just — that's sort of a math thing mostly. And then Luis, maybe you'll talk a bit about how bookings are trending near term.

Luis von Ahn Arellano executive
#10

Yes. I mean, we began the year with a really clear plan to put better end user growth first. And we've — and reaccelerating user growth is extremely difficult, and we're quite proud that we've managed to do it. But we also said at the start of the year, look, our bookings are going to land inside this range of roughly 10% to 12% year-over-year growth. We're at the higher end of that. We feel pretty good about it. But we simply haven't been focusing all that much on monetization; what we have done on monetization this year, though, is try to invent new approaches to monetizing that line up better with DAU. And the good news is that we've found a number of them. One is longer free trials, as you mentioned. There are others, for instance, monetizing the Avatar suits like we discussed, which does really well with IAPs; and there are just a number of other things we're going to be doing, and we're gradually starting to monetize more in that way. Another thing we're really excited about is that Chess is completely under monetized. We're going to be monetizing Chess. We're very proud. We just reached 10 million daily active users in Chess. So a relatively significant share of our users are now doing Chess. And we think there's a lot of low-hanging fruit there.

Ryan MacDonald analyst
#11

Luis, I was hoping you might be able to give us a bit more detail on those first few days of testing Video Call within Super Duolingo — what are you seeing when it comes to user engagement, or even conversion? And then how did that shape your thinking on the choice to sunset Duolingo Max here?

Luis von Ahn Arellano executive
#12

Yes. That's a great question. I mean, when we launched Duolingo Max, the higher subscription tier, we had to do it because we had certain AI features that were simply very expensive to serve. I mean, serving those AI features cost a lot. We did say at the time that if we could get the cost down, we might place these features in other tiers. What happened was that the premium feature of Duolingo Max, which was Video Call, when we first launched it, cost roughly $0.30 per call, basically for AI inference; we've gotten that cost down to well under $0.01 per call, mainly by shifting to open source models or open weight models. So we feel really good about that, and it let us put Video Call into Super. And that was actually a really good thing. When we A/B tested it, it's good for revenue, but it also lets us give Video Call to about 10x as many users, since we simply have far more Super subscribers than Max subscribers. And that's good because it means 10x as many people are learning better, they'll tell their friends, et cetera, et cetera. So we just see this as very positive. And because of that, and when we tested just — we actually just tested a sunsetting of Max, it was very clear that sunsetting Max was the right decision; so we've begun doing that; we don't think — in the short term, we don't think this is going to have a major bookings impact. It's slightly positive. It's not a major bookings impact. But over the long term, we think this is really going to help us because more and more people are going to get Super. I mean, this is one of the main reasons people are going to get Super; so we feel pretty good about that.

Gilian Munson executive
#13

Yes, and one thing that stands out about how we've handled this is that we've reduced our Max exposure quite a bit. So most of our bookings at this point don't come from Max. It's in the single digits for both paid subscriber count and bookings. So it has turned into a very minor item for us to manage, and obviously we demonstrated this year that we're able to handle these things pretty effectively.

Ryan MacDonald analyst
#14

And Gillian, could you perhaps walk through the assumptions around churn as you move through this process? And will you roll this out to existing Max subscribers all at once, or will it happen gradually at renewal?

Gilian Munson executive
#15

Essentially, we've already halted sales of Max to new users. Max individual will shift to Super Family at a comparable price. That means those subscribers gain the ability to add friends and family while keeping all the great features they have in Max as they move to Super. That then leaves us with just Max Family, which is what we'll finish the year on — the smaller share of bookings and paid users that I referenced earlier. So we've really brought that exposure down. And again, closer to what Luis was saying, as we test this, it turns out to be net positive for us; over the long term; near term, we work through it, but as you've seen, we've substantially reduced the exposure there, and we feel very good about how we can handle it.

Shweta Khajuria analyst
#16

Could I ask 2 things, please? First, a follow-up about bookings growth: how do you view that gap between DAU growth and bookings growth as we consider bookings growth catching up — how should we think about that looking toward '27? And the second is about gross margins. Gillian, is the 72% you guided to a solid run rate now that the Video Call transition is going to happen? We think about costs declining, yet at the same time usage for Video rising — is that a reasonable margin run rate to assume?

Luis von Ahn Arellano executive
#17

Yes. On the gross margin question, I'll let Gillian respond. As for bookings growth, our users monetize over time. Usually, when someone is a new user, there's a certain likelihood they will sign up and pay for the subscription. But for many of them, monetization sometimes takes years. So we're simply quite pleased that we now have a considerably bigger user base — an active user base — that will eventually try to monetize. We aren't providing guidance for '27 at this point. What I think we're saying, though, is that we feel fairly good about our bookings growth, and we're beginning to focus more on monetization now that we've really succeeded in reaccelerating DAU growth.

Gilian Munson executive
#18

Yes. On gross margin — clearly, we're not providing guidance for next year, but even so, if you look at this year, we expected gross margin to come in roughly 2 points lower. AI expenses haven't been as large as we anticipated, which is great, and we're still rolling out AI features — things like Video Call — to every one of our paid users before the year is over. So we feel quite good about how we can handle that. I'd say our desire to deliver AI-based features to our customer base is extremely high. We believe there are genuinely creative things we can do with it. So, similar to what you saw this year, we'll steer this toward an overall plan, but once more, no guidance for next year. I'd simply say we have an appetite to give customers as much as we possibly can, and you've watched us do exactly that this year.

Bryan Smilek analyst
#19

I've got just 2 quick questions. In the prior quarter, we discussed CURR being at record highs, and retention is improving among newer cohorts. I'm wondering if you have any fresh details on that—particularly since DAU growth is speeding up, how would the overall curve look on a cohort basis? Also, Gillian, quickly on paid marketing: I think it's up nicely versus a year ago as you get better efficacy. Could you discuss the LTV to CAC ratios you're observing and where marketing dollars are going across channels?

Luis von Ahn Arellano executive
#20

Sure. I can speak a bit about CURR. We brought it up last quarter—it's a metric we take real pride in. It's the key metric we optimize to grow daily active users and boost app retention. We'd rather not start guiding CURR or providing CURR updates each quarter; so no precise figures from us, but it essentially continues rising. And most of the DAU growth you're seeing stems from gains in CURR.

Gilian Munson executive
#21

Now, regarding paid marketing — that’s been a genuine success story for our team this year. To give you some context, paid marketing accounts for roughly 60% of the marketing spend we classify as noncomp, and that comes to a bit more than $100 million. That spend is profitable for us, though we still think there’s additional room to improve the LTV CACs we monitor for the business. As for where those dollars go, the allocation is quite broad-based, and we shift it around pretty actively depending on what’s generating solid returns for us. The team keeps doing that very effectively. So I’d say the net takeaway on paid for Duolingo is that it will likely never represent the majority of how we go to market — that simply isn’t how we think about it — but it’s a very useful tool, our team has performed really well, and we feel returns are improving and that we’re getting smarter about it. So it added to our results in 2026 and probably will again in ’27.

Luis von Ahn Arellano executive
#22

Yes. I’d say internally, becoming genuinely smart about performance marketing is among the largest wins we’ve had this year, and that makes me very happy.

Mark Mahaney analyst
#23

Let's try 2 questions. As you roll out this Video Call feature to more of the Super Duolingo subscriber base, what have you seen so far regarding the boost to engagement, retention, or satisfaction? My assumption is that the impact has been material. It's partly why, as costs come down, you're expanding it. But what does the data actually show so far? Is it too early to tell? And second, I know you mentioned Chess earlier, so what about my second favorite topic, math? What sort of user uptake are you seeing — I couldn't find any figures in here, though perhaps your press release covers it — and could there be a chance to run a standalone or bundled pricing move as well?

Luis von Ahn Arellano executive
#24

Thanks, and I appreciate both of those questions. On Video Call — well, honestly, our confidence in Video Call is high; it's a capability that genuinely drives this conversation, and we're watching its uptake keep climbing. The team simply continues to improve it. One metric we use is simply how many words each active user speaks—not in practices or Video Call, but in Duolingo practices—and that figure continues to climb. I believe it has roughly doubled over the past year or so, which makes us feel quite good. Another point: we know that when people try Video Call during the free trial, for particular kinds of Video Call sessions that we adapted to the user's difficulty level, they become more inclined to sign up and actually pay for a subscription. So we feel really good about it—this feature just keeps improving. On the math side — honestly, this is the subject I enjoy discussing most. Math is something I genuinely care about, and I'm a big fan of what we've built there; it continues to improve steadily. The potential here is enormous as well, since roughly one billion people worldwide are studying math. That's why I see so much opportunity. Its growth pace is roughly comparable to that of languages. I'm glad it's expanding, though I'd like to see it eventually match Chess's trajectory, since Chess is climbing more quickly. So our focus will remain on refining the teaching tool further. A discovery we made, incidentally—and it isn’t a particularly difficult lesson—is that individuals would rather engage in the game of Chess than study mathematics.

Mark Mahaney analyst
#25

Both of those are fine.

Luis von Ahn Arellano executive
#26

Honestly, I'd say the same — both work well for me, but hey.

Omar Dessouky analyst
#27

I'd like to go a bit deeper on the gaming partnerships. You mentioned Brawl Stars and Eggy Party, and you said those brought in millions of new users. So how scalable is that approach? Have you reached a meaningful share of, say, the top 100 mobile game developers? And what's the economic incentive for a game publisher to work with you, especially since you're sort of competing for the same consumer attention? Then I have a follow-up in the same vein.

Luis von Ahn Arellano executive
#28

Yes. We're quite pleased with these partnerships. Broadly speaking, in many cases people approach us rather than us having to approach them. Historically, folks have really wanted to partner with us, so we're very happy. We expect to keep doing some of these. That said, it's not as if there are hundreds of such partners we'll sign. We see it as more occasional. This will likely happen a couple of times a year, and they do bring users, which makes us very happy. But the majority of our events, if you look at the plan for next year, are first-party events; those are our own initiatives — we'll do things like Avatar suits and similar. As for what the game publishers get out of it, no money changes hands in these partnerships; we didn't pay them, and they didn't pay us. What we've found with these partnerships is that both sides win. They gain a bunch of users, and we gain a bunch of users. So it doesn't feel like we're competing for that; generally, the incentive is simply cross-promotion, and for certain games — certainly true for Supercell and Brawl Stars — they do a lot to keep their game fresh. One thing they did when they partnered with us is that for a period of a few weeks, the entire game was teamed with Duolingo, and the players like this. It was a very positive thing in their community. So that's the incentive for doing it. They also gained a bunch of active users and revenue.

Omar Dessouky analyst
#29

All right. Then as a follow-up: does this tie in at all with what you're doing on the performance marketing side? Since, as everyone knows, mobile games are a huge force behind performance marketing across mobile gaming apps generally. So do you see this approach as somewhat complementary to that? Are you aiming to reach a larger slice of the gaming audience and pull those users in via performance marketing, and use this experience toward a — how should we think about that moving forward for you? It's a fairly large audience, isn't it.

Luis von Ahn Arellano executive
#30

Yes. I believe the two are connected. Because you're right that many performance marketing ads are game ads. Definitely the ads we display inside our product too. So yes, we see them as connected. During the period when these partnerships were running, we actually saw our performance marketing gain from the partnership. So that is something we observe. I don't think we yet have a full grasp of how these 2 relate, but I do believe each adds to the other.

Ron Josey analyst
#31

This is James Michael speaking in place of Ron. Since trials each week increased more quickly than staffing levels during the past year, and Green Machine has produced 2 quarters of acceleration, could you give us additional detail on the fundamental shifts in how the testing process is structured? And what further throughput might be possible from AI handling more of the experiment design and analysis?

Luis von Ahn Arellano executive
#32

Yes, that makes us quite pleased. Looking at things now, the count of experiments we conducted — say, during the most recent month — is roughly 90% higher year-over-year versus last year. Meanwhile our headcount rose around 10%, so on a per-capita basis this represents a substantial gain. Much of that — though not all — stems from AI. What we're observing is that it took a while for AI's promise in coding to translate into real speed gains, but that is happening now. And we're seeing not only a much larger experiment count, which I expect will keep climbing higher and higher, but also that latency — the gap from an idea to having it in front of users — is far shorter today. That lets us iterate heavily. I think this fits extremely well with how we improve, since we simply run a huge number of A/B/B tests. What's great about AI is that it excels at these smaller modifications. With bigger changes, it often still kind of breaks things. For the smaller items, though, it performs quite well, and we mostly rely on hundreds upon hundreds of tiny modifications. So we find it very compatible with how we work; that's why we feel fairly confident in our capacity to keep growing both our DAUs through retention gains and our monetization, because we're going to — my sense is next to you. I don't know the precise quantity, but we're going to run a hell of a lot more experiments next year than we did this year.

Analyst analyst
#33

That's helpful. One quick follow-up on it. During the previous quarter, you noted a shorter attention span and tech shorter < >. Has anything changed on that front? And are you observing down-man specifically from session frequency in addition to retention?

Luis von Ahn Arellano executive
#34

Yes. We're working on it — this is something we're truly working on. None of that are we disclosing, but we are actively — we now have teams staffed to try to increase engagement across the board, and about that work we feel pretty good.

Justin Patterson analyst
#35

Great. Well, first off, nice work on the DAU gains this year — you've clearly made a lot of headway. I suppose as we move out of this year and into '27, how are you weighing the appropriate amount of friction and steps to add back into the app and steps toward monetization, so that the lessons of 2025 don't get repeated? And I'll have one follow-up.

Luis von Ahn Arellano executive
#36

That's a great question. I mean, we're — inside the company we have a strict guardrail not to degrade the free user experience anymore. On the positive side, we dedicated this year to determining how to generate revenue while keeping the free user experience intact. I think we've identified several levers that are effective and that we feel quite confident about. That said, yes — this has been a learning experience. We have no intention of harming the free user experience again. I don't expect that to occur.

Justin Patterson analyst
#37

Understood. And then considering the choice to sunset Max this way—moving Max subscribers over to the Super Family plan—could you explain why that's the correct move rather than the individual family plan? Sorry, I mean the individual Super plan. And that opens up the possibility for more paid sharing, so to speak—friends perhaps sharing with each other—so I'd like to hear how you're thinking about the guardrails there. So these are genuinely net new people rather than some canceling and simply joining a friend's plan.

Luis von Ahn Arellano executive
#38

Yeah, that's a fair question. What I'd say first is that right now, despite the family plan being quite a strong value relative to the individual one—even with that being true, we're fond of it since its retention is extraordinarily high. So as time goes on, it just—we're happy with it. Retention is genuinely strong. And it stands to reason that retention is genuinely strong. When you've got people on it, as long as one of them keeps using it, they sort of continue paying for the plan. So that's the reason we're fond of it. The rationale behind shifting Max individual to Super Family was that the two cost about the same; that let us avoid something like a onetime drop in bookings from these subscribers. And from their perspective, we believe it's a solid deal. What I mean is, the features are essentially identical—you just get to invite 5 more people. And that will occur. I'm certain there's a certain degree of cannibalization, but we aren't really observing it in the numbers. It isn't material enough for us to spot anything like, oh, this is actually harmful to us, because we've been A/B testing this for a short while; and we simply think it looks good.

Gilian Munson executive
#39

And Justin, I think that really speaks to how confident we are and how big the market opportunity is here. Being able to do that — and we genuinely aren't worried that we'll hurt our market opportunity. We remain extremely bullish about what lies ahead for us and the billion learners we believe we can reach; that's going to give us a much bigger business over time.

Alec Brondolo analyst
#40

I think the other shift this year for touring, apart from Max being deprecated, is the light tier. I've been noticing lights here. In my own app testing during the third quarter and the start of the fourth quarter, I've been seeing the light tier fairly often, sort of merchandised to me. So any thoughts on what you expect for light penetration next year? And how the advertising experience at Duolingo stands right now would be great.

Luis von Ahn Arellano executive
#41

Okay. So it's light on that. We're testing it. We haven't pushed it very far, and we may try pushing it more. I mean, we're mainly driven by our EBITDA results. And I think our EBITDA for light just hasn't been home runs. There are probably some wins there, but they just haven't been home runs. So I don't actually know what will happen with light. It's a much cheaper plan that also has ads. I just don't know what will happen. It may be that it ends up only being served in certain regions, probably regions with lower GDP per capita, kind of. But I don't actually know. Right now, this is pretty small, and we just haven't seen it be a major home run. What is there is still being tested. On ads, we feel very good about ads; on the team, we now have an actual serious team working on our own ads business. So we feel like next year our ads business is going to do well; and just to remind you about our ads business, historically we haven't invested much in our own ads. And so what we did is we just used these ad networks with kind of the lowest common denominator of ads. And what we're doing now is switching those to much higher quality ads. Now that takes some time to do; in part we're doing it through direct sales, we now have a direct sales team. It's not a very large direct sales team, but it's a very good direct sales team. But we're doing some of that; and also, we're just switching the ads to higher-quality ads; so the transition is going to take some time. But once we're done with the transition, you're just going to see significantly higher quality ads on Duolingo with kind of more well-known brands and higher quality brands and in addition, that pay us more. I think over the long term, we feel actually pretty excited about our ads business.

Gilian Munson executive
#42

During the quarter, the team achieved some genuinely encouraging progress. Looking at the quarter as a whole, the ads business was more or less flat, yet in fact it finished that period growing at a nice pace. So our view is that the team is doing solid work on that front.

Andrew Boone analyst
#43

I'd like to start by building on what Justin asked. When we look ahead to 2027, you've obviously moved from monetization into engagement. As we return to a monetization format, should we anticipate a shift in how you're pursuing that? Would that mean different in-app purchases? Are there certain things that are now off limits? Is this a restarting of the engine that was previously up and running? Or will you approach this in a different way? Then my second question is another follow-up: as we consider AI unlocking Duolingo broadly, Luis, is AI's biggest needle mover internal? Is it new products that become available? And what happens as we've seen AI deflation come through the model? What does that open up for you regarding internal processes that could become available in maybe a year or two years, as we've seen happen with video calls?

Luis von Ahn Arellano executive
#44

Great. Regarding monetization, I mean, I'm not sure precisely how to respond to your question about whether it's a restart. Is it — I mean, we've been focused on monetization throughout this year; it's simply that we instructed the monetization team, look, you can't degrade the free user experience any further. We need to find new approaches to monetizing; and they have. And that's performing really well. What you'll notice about how we operate is really no different from our approach all along: we'll keep running thousands of A/B/B tests aimed at improving monetization. What we won't do is make the free experience harder to use, but plenty of other levers are in play. Yes, IAPs are part of it. Yes, better ads are part of it. Beyond that, there are further ways to draw people into subscribing. Video Call is a strong example — for a very long stretch, Video Call existed only in Max. With it now part of Super, we're far more comfortable promoting it regularly, since the price point is much lower. The truth is that Max had been a fairly costly plan. So overall, we're feeling good about it. That said, the kind of work they'll be doing mostly consists of plenty of A/B tests.

Andrew Boone analyst
#45

Sorry to cut in, but does that mean conversion rates won't get better going forward? Because then you'd have to grow your user base without any extra friction, right?

Luis von Ahn Arellano executive
#46

I can't answer that for you — whether we'll manage to improve them or not. Of course, improving conversion rates is something we want. I don't know. I think certain things — a few of the things we've done this year have boosted conversion rates; for instance, longer free trials do improve conversion rates. And by the way, right now we aren't giving longer free trials to all users. We've sort of provided them to certain users. We intend to extend that further. So I do think there's some lift in conversion rates. How much, I don't know. But one thing that's certain is that you won't see us squeezing users. That's not something you'll see from us.

Gilian Munson executive
#47

Yes. And Andrew, I believe you already noticed during the quarter that our bookings growth picked up speed compared with the prior quarter, and I think what you're seeing is that we're being cautious in how we approach all of that. One point we've discussed with all of you is that the top of the funnel carries a lot of weight. The top of the funnel converts at a multiple of what the core base does. So that explains why you've watched us focus on areas such as paid marketing, along with certain outreach efforts expanding into new places and new products. And I think that's one piece of the puzzle on an aggregate basis.

Luis von Ahn Arellano executive
#48

Then the second part of what you asked about was AI. There are two areas — and you named the two areas where AI genuinely provides a lot of help. One is simply internal productivity, and the other is AI features within the product. We're quite pleased about both. On the internal productivity side, as I mentioned earlier on this call, we're seeing our team able to ship A/B/B tests far more quickly. And we're reaching a point where the simpler A/B/B tests are now being shipped by non-engineers. So someone who is a product manager or a designer can now take something all the way from the idea to it being in the app. They can do that on their own now, without an engineer's assistance. That doesn't mean engineers aren't needed. We need engineers for all sorts of things, including the system — we built an internal system that lets us do this. We keep trying to improve that system more and more. So we definitely have a major need for engineers. But the fact is that a significantly larger number of people are able to run their own A/B/B tests. So I don't know the precise increase in the number of A/B tests we'll run next year, but it will be multiples of what we ran this year. I don't know whether it will be 3x. I don't know whether it will be 10x. I don't know, because all of this stuff is changing, but it will be multiples. So I'm very happy about that. And that, I think, is one of the things we feel bullish about. On the product side, you'll simply see us put AI in a lot more places. We're working on a number of features, particularly practicing speaking, that aren't just Video Call, and you'll see those there. Beyond that, you'll simply see improvements in how fast we can ship new teaching modules, let's call them. For instance, we're just going to teach a lot more Chess, a lot more content. We'll just be able — we're able to ship it way faster now.

Alex Sklar analyst
#49

A two-part question about gross margin. First, for you, Gillian: besides what you laid out for the fourth quarter, will Max being sunset have any effect on gross margin? Then second, for you, Luis: regarding the share of direct payments made outside the app store, has anything changed in how you view that opportunity — whether because personal agents are rising and pushing more traffic straight to you, or simply because of more market testing?

Gilian Munson executive
#50

So, Alex, our fourth-quarter guidance, which implies gross margin declining roughly a point-ish, does include the Max transition we anticipate. As you've watched our team do throughout this year, we always keep an eye on becoming more efficient with our AI spend. That said, we also want more AI in the product and to balance that against our bottom line; so we do believe you'll see gross margin step down sequentially, yet it remains a very attractive figure, and we'll be able to add considerably more AI content to the product while also completing the Max transition.

Luis von Ahn Arellano executive
#51

On the direct payments question — yes. For a very long time, at least inside our apps (we have both an iPhone app and an Android app, plus our website), direct payments could only happen through the website. Now, in some regions and for some users, we're able to accept direct payments within the apps themselves. We apply it in select situations, and one area where we lean on it fairly heavily is performance marketing. With performance marketing, we bring users in through a website first and then encourage them to install the app — that's fairly standard practice — so you'll see us doing some of that. Honestly, just going by how many folks we've got assigned to it, I'm not sure it'll be a big contributor in the coming year. Sure, we apply them in various spots, but we aren't at a point right now where we feel everything has to move over to direct payments. That's not something we're planning at present.

Alex Sklar analyst
#52

Understood. And regarding that record gross margin for Q3, wouldn't you credit some portion of it to the rising share of traffic you're driving straight to your own platform?

Gilian Munson executive
#53

No. What drove the gross margin improvement was primarily stronger revenue combined with lower AI expenses than we had anticipated incurring.

Ross Sandler analyst
#54

Luis, let's step back and look at the big picture when it comes to AI. All of us can see that AI is moving forward at an enormous pace. The models continue to improve. Latency continues to get quicker. Meta even rolled out a rather sassy avatar not long ago, following in your footsteps. But this is what I want to ask: some academic research indicates that one-on-one tutoring is more or less one of the finest methods for anyone to pick up a new subject. So earlier you spoke a little about how you'll bring some of these more native AI interactive elements into the app. At present it's been somewhat of an add-on next to video. But I'm curious about your long-term thinking — could these turn into something far more central, rather than the gamified tap-and-learn kind of use case we began with at Duolingo? And how does that shift, I suppose, the business model for Duolingo as you bring more of that in?

Luis von Ahn Arellano executive
#55

Yes. That's a great question, and it's something we think about a lot. We know that today — a few years back, maybe 4 or 5 years ago, we conducted a bunch of research showing Duolingo was about as good as a classroom for language learning. We've improved to the point where I'm fairly sure we're better than a classroom now. Right now, though, we don't yet match a single student working with one human instructor. No system has gotten to that point—a strong individual human teacher remains ahead of our current performance. We expect that within the coming few years we can reach parity with one-on-one human tutoring, simply because we observe the gains we make annually. The key point about individual human tutors is that their effectiveness stays flat—they're roughly at their ceiling—while we keep getting better. That's why we believe we'll arrive there. One key point deserves emphasis here: although individual human tutoring works extremely well for instruction, the reality is that the vast majority of people simply aren't interested in it. Sure, a tiny share of folks are open to it, yet most would rather not — and that holds true even when the service costs nothing; the simple fact is that, set beside TikTok, Instagram, or similar apps on a phone, the experience just isn't engaging enough. It simply doesn't hold much appeal. Our goal is to deliver something with the same effectiveness as a one-on-one human tutor, and we believe that's achievable, though it won't resemble a one-on-one human tutor. It's still going to feel gamified in the end, because the most crucial thing — even though one-on-one human tutors are highly efficient and effective — the most crucial thing is that the main failure mode is simply people dropping off. You can have a very effective one-on-one human tutor, but people will just be like, yes, I'm going to stop going to that. The direction we're taking is that it will keep that game-like quality, though you're going to notice considerably more artificial intelligence involved as we aim to engage you right at your own level. Incidentally, the thing that makes individual human instructors superior to a group setting — or to our current offering — is their strong ability to tailor themselves to each learner. They simply — they — and they'll even pick up on cues such as your face while they're talking with you, noticing when you seem to be, say, puzzled. They pick up on things without you needing to spell them out, which is exactly why they perform so well. In my view, this technology keeps improving, and we'll manage to achieve that. It won't happen overnight, but I believe we'll reach that point.

Gilian Munson executive
#56

But then, Ross, just look at how many people use Duolingo and the database behind it — that puts us in a spot where we can deliver the best there is on the market. I think that's a genuinely powerful asset for us. The extraordinary volume of people who come to us daily, and using that data to figure out precisely the right way to teach them.

Luis von Ahn Arellano executive
#57

Yeah, that’s a really strong point. When it comes to data about people learning, I mean, I don’t believe anyone — well, we know, we just — no one possesses the volume of learning data that we have.

Nathaniel Feather analyst
#58

I've got two. First off, you mentioned putting more emphasis on monetization next year compared with this year—what does that look like inside the company? Does it mean more staff assigned to monetization work? Does it mean the bar for what you'll approve shifts? How are we supposed to interpret what that really involves? And then second, given how well the Chess course has done, how are you viewing expansion into new courses, both regarding the timeline for when you might push on that more aggressively? And also the kinds of subjects you believe could work here might differ from what you tried before.

Luis von Ahn Arellano executive
#59

Both excellent questions. What does focusing more on monetization mean? Well, this year was unusual because we told the entire company, look—what truly matters is speeding up DAU growth. There's a massive opportunity in front of us. We need to work out how to speed up user growth. And honestly, speeding up user growth is genuinely difficult; so hey, let's tackle it. So we set a bonus for the entire company, and we said, let's go speed up user growth. Next year looks a bit different. What we're telling everyone is that we keep our focus on user growth since the opportunity ahead is so large. But you know what else counts, monetization; so I think you'll simply see additional A/B/B tests tied to monetization, and that's roughly how we'll run things next year; as for new courses, we aren't announcing any new courses right now. What I will say is that when we added Chess, AI contributed quite a lot; and we managed to add Chess from the idea stage to being live in the app in roughly 9 months. What I will say is that we know we can now do that even quicker.

Arvind Ramnani analyst
#60

All the same, though — yeah, I'd like to ask specifically about inference expenses and DAUs, correct? Your daily active users have risen nicely, roughly 30%. How does that affect the cost for each DAU? Basically — is the per-year cost declining more quickly than AI consumption per user is climbing?

Luis von Ahn Arellano executive
#61

Yeah, that's what we're observing. What I mean is we aim to add plenty — we've already been adding far more AI into the product, and we're watching cost per DAU drop substantially. The biggest factor is that, for most of the people using our product, open weight models perform about as well as frontier year models. A few narrow situations still call for a frontier model. Yet for most of the people using our product, open weight models are more than sufficient; so any time we can move from a frontier model over to an open weight model, the savings are quite large, and we manage to do that fairly often; so our team has done a genuinely excellent job of handling that responsibly.

Arvind Ramnani analyst
#62

Yes. No, absolutely. And I think you're sort of leading the way on this now — it looks like that's roughly what the more experienced users, or the heavier service users, of AI do: they begin with the frontier model and in time they migrate. Especially if you have, you know, a certain level of sophistication; and I suppose the other piece is that you'd want to put additional AI into the free experience so DAU growth gets driven; and I think you mentioned one of the main reasons the cost of video fell by something like 95%; so what I want to ask is, at what point do inference costs for free users stop making sense, right? I mean, inference can be delivered quite cheaply; but there's a breakpoint where inference cost per free user truly stops making any sense. And then you say, hey, we're going to begin offering this to the free user.

Luis von Ahn Arellano executive
#63

Yes. I suppose the way I'd respond to your question is that next year we are going to spend some amount on inference cost for free users. It isn't a ton, but we will spend some on it, because we do want to give free users some of these AI features. But you'd describe it as single-digit millions that we'll spend next year on inference costs for free users, on roughly live inference costs.

Arvind Ramnani analyst
#64

Yes. And sorry, this is the last question on the topic. Are you still, like — and I'm not asking about the next quarter, but are you only a few years out from your business getting big enough that you'd essentially own some of the compute? Or will you — how are you getting at this, right? Are you going through the AI cloud? Are you going through the inference providers? Are you handling it in-house? Like, how are you getting the inference or the compute to do this?

Luis von Ahn Arellano executive
#65

Yes. I mean, it's all of the above. I'd say inference providers are the majority at the moment. But we're trying some internally. So it's kind of all of the above, but we'll — we'll do whatever makes sense for the business. Yes.

Arvind Ramnani analyst
#66

Yeah, traditional A/B/B testing — they've always been really sharp about that.

Gilian Munson executive
#67

Right. The team is quite inventive here and has handled it extremely well.

Ralph Schackart analyst
#68

I joined the call late, so forgive me if this was already covered. During the call, though, you mentioned that you'll be running a number of extra tests in 2027, and they cited something like 90% year-over-year testing growth year-to-date. My assumption is that this is on the product side, but could you clarify whether all of it is on the monetization side? Maybe the bigger question, though, is how can you run these tests next year at this scale? And then also keep the focus on monetization. Do those experiments need to come down by, I don't know, a rate of 50% to 45% year-over-year growth? Could you take us through how you can keep these experiments going while also driving monetization next year?

Luis von Ahn Arellano executive
#69

Yeah, so the way we make the product better is by focusing on 3 main areas; we improve what we refer to as user growth. That typically means boosting retention of the product, which lifts daily active users; we improve monetization, and then we improve how well we teach. Across all — across these 3 buckets, our approach is basically to run a huge number of A/B tests in each 1 of them, with the aim of improving one of these things. So this specific A/B test is designed to improve monetization. The goal of this specific A/B test is improving teaching, among other things. And we conduct many of these. As we said, we're currently running roughly 90% more than, say, last month—compared with a year ago, we did 90% more A/B tests. Our expectation is that this will continue rising; so broadly, we'll be in a position—for monetization, and also for better teaching, and for daily active users too—to carry out far more experiments in the coming year. It's hard to say how many more, since that hinges on various factors—how much our processes get better, and so on. But we're reaching a stage where the constraint is shifting: you begin to hit cookie-related constraints, which aren't what we traditionally viewed as constraints. Take ideas, for instance—these days a constraint is coming up with ideas. I mean, even generating concepts for what to test has turned into a constraint for us. Another constraint is, say, one capability we now have—as we noted in an earlier response—is enabling people who aren't engineers to actually launch an A/B/B test. Take a product manager or a designer — they can actually do it now. We created an in-house tool that lets a PM make a full change on their own, with no engineer involved; these aren't big changes, just small ones. But we can do that, and then another bottleneck shows up: suppose that person is a product manager, they introduce a bug into the app, and who fixes it? Because I doubt they'd know how to fix it. So this is the situation — our processes are going to have to change. We're right in the thick of it, and honestly, on a few of those items I can't yet tell you what the scenario looks like or how we'll handle it. What I can say is that we'll be able to run some whole-number multiple of the A/B tests we did this year.

Luis von Ahn Arellano executive
#70

I appreciate everyone joining us, and I appreciate the excellent questions. And in particular, my thanks go out to the team for the truly outstanding effort so far in this year. Our results make us quite proud.

Read the full transcript via the API

You're viewing the first half of this call. Get the complete Duolingo, Inc. transcript - plus 132,000+ transcripts from 7,000+ companies, speaker segments and full-text search - through the EarningsAPI REST API or hosted MCP server.

Get an API key View API docs →

For developers and AI pipelines

Programmatic access to Duolingo, Inc. earnings transcripts and 132,000+ others is available through the EarningsAPI REST API and the hosted MCP server. Quarterly plans from $162 - full transcripts, speaker segments, full-text search, and the /api/v1/transcripts/recent polling endpoint for ETL pipelines.