Home / Transcripts / Duolingo, Inc. (DUOL) · August 5, 2026

Duolingo, Inc. (DUOL) Earnings Call Transcript

August 5, 2026

NASDAQ US Consumer Discretionary Diversified Consumer Services earnings 45 min

Earnings Call Speaker Segments

Deborah Belevan executive
#1

Hello, everybody, and thank you for joining Duolingo's Q2 earnings webcast. Earlier today, once the market had closed, we put out the shareholder letter for this quarter, and you can access a copy over on our IR site, investors.duolingo.com. Joining us on this call are Luis von Ahn, who is our Co-founder and CEO, along with Gillian Munson, our CFO. Each of them will give prepared remarks first, and after that we'll take questions. If you're an analyst and would like to ask something, please use the raise hand function. I'd like to point out that this call is being recorded, and at the moment all participants are in listen-only mode. Before we get started, I want to mention that we will be making forward-looking statements about future events and financial performance. Such statements carry risks and uncertainties that are described in our SEC filings, and they rest on assumptions that we consider reasonable as of today; we assume no duty to revise them. In addition, we'll cover both GAAP and non-GAAP financial measures. You can locate reconciliations of the two in our earnings materials. We suggest you look at those when assessing how we performed. At this point, I'll hand things over to Luis.

Luis von Ahn executive
#2

Thanks, Debbie, and I appreciate everyone being here. We had a solid second quarter. Daily active users rose 23% compared with the same period a year ago, picking up speed relative to Q1 and landing a bit above what we had anticipated. What we've observed thus far in Q3 gives us reason for optimism. The bulk of that expansion stemmed from our everyday efforts through the initiative we refer to as The Green Machine; we put hundreds of product modifications to the test, gauge how much difference they make, and invest more heavily in the ones that deliver results. Though nearly all of these tweaks are minor, their effects build up as time passes. As I explain in our shareholder letter, this is likewise what is pushing CURR, a metric for user retention, up to its highest level ever. One more standout from the quarter was Streak Revival, a single-instance campaign that took place in June. The concept was straightforward: Offer learners who had broken their longest streak an opportunity to win it back by finishing three lessons. Over 15 million learners got their streaks back. What's especially heartening is that these users are demonstrating stronger retention than a standard re-engaged cohort. We returned learners to a product that continues to improve at teaching languages, Chess, Math, and Music, and they are remaining. We remain in the early stages of carrying out our approach of putting user growth first and teaching more effectively, yet Q2 left us feeling more assured that our direction is correct. With that, I'll hand things over to Gillian.

Gillian Munson executive
#3

Thank you, Luis. Hello, everyone. As Luis mentioned, the second quarter was a solid one; beyond the DAU acceleration he just covered, our top-line results matched what we had anticipated, and profitability came in a bit better than planned. Looking ahead to the rest of 2026, I'd like to restate how we're running this business. We're making intentional investments in areas we think can turn Duolingo into a much bigger company over the long haul. Meanwhile, our team keeps operating with discipline. We're still working toward our full-year bookings and revenue target ranges of 10%-12% bookings growth and 15%-18% revenue growth. Our target adjusted EBITDA outlook has been raised to 26.5%, up from the 25% we laid out at the beginning of the year. When it comes to point estimates to assist with your modeling, please note the following. For the full year, we anticipate bookings growth of about 11% and revenue growth of roughly 16%. Using constant FX rates from our prior call, that bookings growth rate would be roughly 0.5 point higher. On gross margin, we now anticipate finishing the year nearer to 70% versus the 69% we originally projected, as we push more AI content into our products, offset by savings on AI costs. We anticipate adjusted EBITDA of about $320 million, at the roughly 25.5% margin I just referenced, and we expect to produce more than $375 million of free cash flow this year. Although it isn't part of our 2026 guidance, we want you to be aware that the company has a bonus plan that would trigger if Q4 DAU growth reaches 25% or above, with payment occurring during Q1. Because it's presently unclear whether that threshold will be reached, we haven't factored it into our 2026 guidance. If it were reached, we'd anticipate the payout to be about $10 million in cash, possibly more if DAU growth comes in higher. For Q3 specifically, we anticipate bookings of roughly $307 million, or 9% growth. Revenue of $302 million, which represents 11% growth; we anticipate gross margin of 71% and adjusted EBITDA of about $76 million, a margin of 25.2%. Our balance sheet and cash flow potential stay strong; we closed the quarter with $1.3 billion in cash and investments and produced $79 million in free cash flow. We bought back roughly $44 million of stock in the quarter, which brings cumulative repurchases under our authorization to $72 million, or approximately 700,000 shares. Taking everything together, our user momentum is strong. Our business model keeps generating substantial cash flow, our team is executing well in a significant investment year. We stay focused on hitting 100 million DAUs in 2028, and we believe the path to that goal can build a considerably more valuable business for our shareholders. I'll now hand it back to the operator, and we're glad to take your questions.

Wyatt Swanson analyst
#4

Hello, and thank you for asking. I appreciate that.

Luis von Ahn executive
#5

Hey there, Wyatt.

Wyatt Swanson analyst
#6

Hi. Since daily active users are now projected to exceed 20% growth compared with a year earlier during the back half of the year, might you offer some insight into why the bookings outlook for the entire year wasn't lifted further? I understand you're currently testing things, but shouldn't having more people on the platform in theory mean more bookings, even if you aren't prioritizing monetization?

Luis von Ahn executive
#7

Yeah, I believe that's an excellent question. The initial point to keep in mind is that people who use our service don't start paying right away. A portion of them do eventually pay, but it takes time for that to happen since our model is freemium. We do anticipate that greater DAUs will lead to greater revenue, though that will require some time. The second factor is precisely what you mentioned. At the start of this year, we stated that we would stay within this box when it comes to revenue, and we will keep to this box. That's approximately 11% bookings growth compared with a year earlier. The remaining work goes toward growing DAUs and toward teaching more effectively.

Wyatt Swanson analyst
#8

Understood. All right, that's useful. You mentioned that extending free trials boosts engagement as well as monetization. Would you walk through a few of the mechanics behind how that actually functions, moving from 1 month to 2 months? Thanks.

Luis von Ahn executive
#9

Yes. To frame it, the objective our monetization team has for this year is identifying things that generate revenue without conflicting with user growth. In the past, a number of the methods we used to monetize involved introducing friction into the free user product, and that conflicts with user growth. One of the approaches that has performed best is longer free trials, and specifically, the primary one we're testing is — historically at Duolingo, the free trial on Duolingo, when we say, "Hey, you can try it for free," has been 7 days. We give you 7 days at no cost. That's what has happened historically. We are now moving most of our free trials, though not quite all of them yet, to a one-month free trial. How that works is essentially far more people choose to take that trial since it's simply a better deal. As a result, we ended up with a bigger number of people actually converting to payers. Another good part is that the moment they accept going on the free trial, the experience improves right away because we switch off energy, we switch off the ads. This actually raises daily active users too. We're very happy with it, and it's gaining a lot of good traction.

Wyatt Swanson analyst
#10

Cool. Thanks, guys.

Luis von Ahn executive
#11

Thank you.

Luis von Ahn executive
#12

Hey there, Andrew.

Andrew Boone analyst
#13

Hello, everyone. We really appreciate you fielding our questions. I'd like to ask about your position when it comes to influencers. How should we view overseas marketing along with the shifts you all pointed out in that letter? Then, thinking about the U.S., Luis, you grasped the quarter's strengths and sort of brought users back. Could you discuss the top of funnel, though? What's your read on the trends there, plus the larger chance of drawing in people who might be brand new to Duolingo? Thanks.

Luis von Ahn executive
#14

Yeah. To give some background, I ought to note that our growth has mostly come organically, driven by word of mouth, thanks to our marketing team. Actually, throughout the earliest years, word of mouth made it entirely organic, at 100%. We brought in essentially a single marketing channel, social — meaning our very own accounts on social media. That continues to perform strongly; we're receiving over a billion impressions each quarter from those accounts of ours, which is amazing. Our marketing team is genuinely beginning to broaden into additional key instruments. Creators are one of these, collections of creators; that's actually going quite nicely, particularly in specific nations. Nations such as China, Indonesia, and India. Roughly 2/3 of all our social media impressions originate from influencers, and we're discovering that we can avoid spending a huge amount of money on this. The good part is that these folks reach audiences unlike ours; this genuinely pulls in plenty of fresh users, and this is something we're quite pleased about; we deploy influencers in varying ways across nations, with some nations getting more than others; that's performing very well. I also want to note that our marketing team has recently grown considerably more advanced regarding performance marketing. Even though organic still drives most of our growth, we're gaining solid momentum with both influencer usage and performance marketing, and we're observing that at the top of funnel. It's rising. We honestly feel fairly positive regarding our top of funnel. It's rising. You mentioned the U.S. specifically. Over the past quarter, U.S. growth has risen considerably, and that's something we're pleased about. Incidentally, when you examine our growth, our DAU growth is genuinely wide-ranging. Essentially, every region is expanding, and every region is expanding more quickly than previously. Asia remains the quickest-growing, yet the U.S. has risen.

Gillian Munson executive
#15

Andrew, one more point worth making is that the quarter's narrative also includes what happened at the top of the funnel. It improved. That area, as you're aware, has been something we've concentrated on. We've discussed wanting to see improvement in that area. During the quarter, the growth rate at the top of the funnel got better in nearly every region, and that's a tremendous achievement for the team.

Andrew Boone analyst
#16

Thanks.

Nathan Feather analyst
#17

Hello, all. Appreciate you answering the question.

Luis von Ahn executive
#18

Hello, Nathan.

Nathan Feather analyst
#19

From where I sit, regarding how you're making money, I'd like to know what you've discovered up to now about offering Video Call to people who newly subscribe to Super Duolingo, and in what way does that shape what you intend to do with Video Call and Max more broadly moving forward?

Luis von Ahn executive
#20

Yes. Video Call is a fascinating item. This feature is one we're fond of. For learning to converse, for rehearsing conversation, it's a fantastic feature. It genuinely delivers results. Actually, we possess research demonstrating that improved conversation skills come to those who use Video Call; it genuinely delivers results. It's genuinely great. Back when we initially began placing Video Call onto the platform, at the moment of our very first launch of Video Call on the platform, I recall being told by the team handling it, "Alright, we're able to hand this to users, yet handing it to users is going to run roughly $0.30 for each call." That was costly, and that's the reason we chose to place it behind our priciest plan, namely Duolingo Max. At that time we stated, "Should we manage to lower what this costs, we're going to test it in other places." Our goal is to deliver Video Call to the greatest number of people we can, since it genuinely assists with language learning. The fortunate part is that, thanks to an enormous amount of truly strenuous effort, we've managed to lower what Video Call costs. These days it sits below $0.01 for each Video Call. What drives that is chiefly a shift in the direction of open source models; doing so is simply far less expensive, and we detect no quality decline there. That pleases us greatly. Thanks to that, we can now hand Video Call to Super Duolingo subscribers. Where things stand at present is that the majority of new Super Duolingo subscribers receive Video Call. Should you purchase Super Duolingo at this moment, Video Call will come to you too; our expectation is that across the coming few months, Video Call will be given by us to subscribers who already exist. Doing so for every one of them isn't something we've quite finished; we anticipate doing that too, and that's genuinely positive since a greater number of people will get to rehearse conversation. That raises the question, what's happening with Duolingo Max? My response is, I don't know yet. A few options exist, and one option might be that while a restricted form of Video Call, such as a capped quantity of Video Calls, goes to Super subscribers, unlimited access goes to Duolingo Max subscribers; that is one option. Another option, honestly, is that Duolingo Max might actually be sunset by us. What I'll say to you is that within the coming couple of quarters we'll have an answer to this, and on top of that, we'll attempt to accomplish this with no revenue lost. This is part of the reason we aren't moving extremely quickly on this, since we're working out how to pull it off while giving up only a little revenue. Once more, the goal on our end is to hand it over to the largest number of users we can.

Nathan Feather analyst
#21

All right, that's quite useful. One item: we've come across reports that you're trying out an ad-supported option. Could you help us reason through whether there's a chance for perhaps a cheaper plan, or something sitting between Super Duolingo and the free version? Looking at it through user segmentation, what exactly are you targeting there?

Luis von Ahn executive
#22

Yes, we're testing that. Super Lite is the name; testing is underway. I truly want to stress the term tested. How Super Lite ultimately turns out, I can't say. Compared with Super Duolingo, it costs less. Roughly 50% of the price, varying by geography. Ads are part of it, so you do see ads; beyond that, essentially, unlimited energy isn't included. What you get is double the energy. That's the concept. Right now, only a modest share of our subscribers are on Super Lite, though one reason is that testing is still at an early stage. We simply aren't promoting it especially well. Naturally, the aim with a lite plan such as this would be: if we're fairly confident you won't purchase Super Duolingo, then we ought to try selling you Super Lite. That's roughly the concept. Over the coming few months, you'll watch us run experiments on that. What will ultimately happen, I don't know; it's something we're trying.

Nathan Feather analyst
#23

All right, that's really useful. Thanks.

Bryan Smilek analyst
#24

Excellent. Appreciate you answering our questions. Good to see you, Luis.

Luis von Ahn executive
#25

Hi, Bryan.

Bryan Smilek analyst
#26

I'm interested. It's great to see CURR hitting a record high. I'd just like to know, can you give us additional detail regarding retention overall broken down by cohort? Do those new free trial folks who use the app every day show stronger engagement patterns? I'm just wondering if there's anything else you could say about retention looking ahead.

Luis von Ahn executive
#27

Yes. As we said in the letter, our retention — user retention — and essentially every measure of it, sits at a record high. The metric we monitor most closely goes by the name CURR, meaning current user retention rate. We're quite pleased that it's at a record high and that over the past year it rose by roughly one percentage point, since small shifts in CURR ultimately translate into fairly big shifts in daily active users down the road, given how strongly it compounds. This is fairly widespread. In essence, you observe it in each region. You observe it among every kind of user. The cause is simply that our product has become stickier. We've introduced quite a few things. It's difficult to single out one, because, once again, we release a fresh version of the app weekly. Each release of the app carries around 350 modifications; it's tough to identify the thing that truly caused it, but on the whole our product has simply gotten stickier. We're quite pleased about that, since that's arguably the finest form of growth you could hope for. It's just pure — it's a superior product.

Bryan Smilek analyst
#28

Great. Thanks; that's really useful; and then, sort of extending what Nathan asked about Duolingo Max in general, more from the speaking side too, could you give us additional detail on overall engagement among intermediate and advanced learners, who in my opinion would probably begin doing more speaking practice inside the app?

Luis von Ahn executive
#29

Yeah. Two points come to mind there. The first is that when someone can use Video Call, engagement looks really strong; as a matter of fact, that feature has recently improved quite a lot. Among the key metrics we track for it is how many words are spoken by DAU who have that access. That chart is lovely to look at, since it slopes fully upward across the past 2 years. Month after month, we do slightly better than the month before at getting you to talk more. In the end, that simply means learners do better and stay involved. That covers Video Call. Beyond that, within the free tier, we're getting people to talk more as well, since there are additional speaking exercises. There are also additional ways to respond to exercises that previously required tapping on the phone, and now a voice can be used instead. We feel quite good about this; it matches exactly what you described. For advanced and intermediate users, this matters considerably more. We view it as something that will, as time goes on, genuinely support monetization, and also word of mouth, since when people learn better, they'll spread the word to friends, and so on. We take a lot of pride in that.

Bryan Smilek analyst
#30

Thank you, Luis.

Luis von Ahn executive
#31

Thank you, Bryan.

Luis von Ahn executive
#32

Hey there, Ryan.

Ryan MacDonald analyst
#33

Luis. Hello, Gillian.

Gillian Munson executive
#34

Hey, Ryan.

Ryan MacDonald analyst
#35

Thanks for taking my questions. I wanted to follow up on the bonus incentive point you raised near the close of your prepared remarks. Clearly, the internal team is pushing toward hitting that 25% figure. How should we interpret what that means for, say, the speed of new experimentation from here through the end of the year? If things accelerate, are there particular features or initiatives you're especially enthusiastic about? Or is the real story simply allowing the adjustments and tests you've already launched to keep maturing in the market and generate their own outcomes?

Luis von Ahn executive
#36

It's a bit of both, honestly. Our testing pace is going to stay high — it's already quite high this year. When you actually count the experiments we release each week, that figure keeps climbing. It's gone up a lot over the past year. In fact, it's grown more quickly than our headcount has. What that tells you is that on a per-person basis, we're getting more experiments out every week. You'll keep seeing that for the remainder of the year. A great deal of experimentation lies ahead. Broadly speaking, that applies to everything we test. Better teaching and growth in daily active users are the areas getting most of our attention, but you'll also see us testing things in monetization and so on. In a sense, it ties back to the bonus, but for the most part it's simply about this year. As an organization, we made the call that the bulk of what our amazing experimentation engine does would go toward one thing: keeping daily active users on an upward path. Our belief, which we've stated repeatedly, is that reaching a company with 100 million daily active users means a far larger business, and that's the destination we're aiming for. Experimentation is something you'll notice. You'll see plenty of it, just as you typically do from us.

Ryan MacDonald analyst
#37

Great; so perhaps regarding Duolingo Max. Clearly, you're currently involved in numerous initiatives tied to Max, including something like Super Duolingo or Video Call, which is being expanded further into Super Duolingo. At this moment, is there any, what I'd describe as, fresh feature product work that might be oriented or directed toward the Duolingo Max tier, which could possibly keep prolonging that tier's lifespan? Or is essentially all of it currently concentrated on experimentation within the lower tier?

Luis von Ahn executive
#38

I won't exclude that possibility. It could turn out that we create something we ultimately place behind Duolingo Max. We're building many features, and some might, due to their expenses or similar reasons, wind up placed behind Duolingo Max; that isn't the objective. The objective is genuinely to attempt to deliver, especially speaking features, since those are the priciest ones to offer. The objective is to make it available to the greatest number of users we can. Once more, we simply hold the view that the more individuals can use this, the more recommendations spread, the more users we gain, and the bigger this enterprise grows. That's the objective. What I suppose I'll add is that isn't what we're attempting to accomplish, though it's possible that within 2 months, I return and mention, "Hey, Duolingo Max in fact received a new feature." That isn't due to any attempt on my part to be secretive toward you. I simply lack knowledge about what will ultimately occur concerning the expense of particular features we're creating.

Ryan MacDonald analyst
#39

Appreciate the clarity. Thanks, Luis.

Gillian Munson executive
#40

Thanks, Ryan.

Luis von Ahn executive
#41

Thanks, Ryan.

Luis von Ahn executive
#42

Shweta, how's it going?

Gillian Munson executive
#43

Hello.

Luis von Ahn executive
#44

Hi.

Shweta Khajuria analyst
#45

Great, thank you. It's nice to see the two of you. I appreciate you answering my questions. My first one concerns users' attention spans. As focus gets shorter, I suppose, what's the trend for engagement and session time throughout the app? And how are you tackling that? My second one is about Math and Music. Would you discuss the product roadmap — how you expect Math and Music to evolve over the year, and how that might add to DAU growth? Thanks.

Luis von Ahn executive
#46

Yes, thank you. Part of what you're asking, I believe, is something we've mentioned before: worldwide, attention spans are shrinking. When you watch how people interact with social media apps, they engage in roughly 10-second bursts. For Duolingo, though, a lesson takes about two minutes. That's an issue we have to deal with, and we're working on it — attempting to extend the minimum session length. We're running tests in that area, though nothing has launched yet. One piece of the problem: if you shorten the minimum session length, that's fine, but you must ensure users return many more times during the day so it compounds. That's more or less what we're experimenting with; we're actively testing it. Regarding Math and Music, we're quite enthusiastic about both. I should note that compared to Chess, even though they each have single-digit millions of DAUs, they're considerably smaller than Chess. We anticipate growth from both, but their contribution to our overall total — we have 60-some million daily active users — even 50% growth there doesn't add all that much. We do expect them to expand. The simplest one to discuss strategy-wise is Math, since we understand it well, or at least better than Music, I'd say. With Math, when we first launched it, I assumed — incorrectly — that we could get an ordinary person to become addicted to learning Math. I was mistaken. I no longer think that's possible. Our approach is really to teach math primarily to those who genuinely need it, essentially K-12 students. We may not sell to schools — that isn't our aim, selling to schools — but at minimum our math user base is mostly under 18. Once we began focusing on that, it clarified many things, and I believe we made far more progress, instead of attempting to make your typical 35-year-old who dislikes math suddenly adore it. I'd love to accomplish that, but I've abandoned that idea. On music, it's somewhat early. We're doing a great deal of work on music. Expect additional comments regarding music possibly in one or two quarters.

Shweta Khajuria analyst
#47

Alright. Appreciate it, Luis. By the time someone's 35, it might be a little too late when it comes to math. If that's not your thing, then that's not your thing. Fine. Thanks a lot.

Luis von Ahn executive
#48

It's not your thing. My whole aim was to make math something people genuinely love; that's my goal.

Shweta Khajuria analyst
#49

Same here. Sounds good. Thanks. Thanks to the two of you.

Luis von Ahn executive
#50

Thanks, Shweta.

Mark Mahaney analyst
#51

Hi there. I'm attempting to join this call. OK. My apologies for that.

Gillian Munson executive
#52

No problem at all.

Mark Mahaney analyst
#53

Let me jump in here. Hi to the two of you. Regarding what was said about the Video Call feature — sorry, Luis, were you indicating that it could be offered to every Duolingo user, or just to those with a Super subscription?

Luis von Ahn executive
#54

Honestly, offering it to every Duolingo user would be my preference. At this moment, though, I don't believe that's something we're able to do.

Mark Mahaney analyst
#55

Alright.

Luis von Ahn executive
#56

What I'm aiming for right now is getting it into the hands of every Super subscriber.

Mark Mahaney analyst
#57

All right. The only thing standing in the way of making it open to every user is simply the price, correct? Or I should say, the cost. If you lower it by an additional 90% in the coming year, then it becomes possible.

Luis von Ahn executive
#58

Yeah, there's one more item that we have to factor in. This ranks among the biggest reasons that drive people to make a purchase.

Mark Mahaney analyst
#59

Yeah.

Luis von Ahn executive
#60

Suppose every user suddenly had access to it — that would reduce the motivation for purchasing, and we'd then need to come up with different reasons to persuade people to buy. There's somewhat of a trade-off in this situation; it involves cost, and it's among the primary factors that drive people to make a purchase.

Mark Mahaney analyst
#61

All right. Now let me turn to a different topic and ask about advertising. My sense is that your stance has shifted somewhat over the past 2 years regarding both ad revenue and ad spend. Where do things stand on that path? Have you observed anything that has made you somewhat more optimistic about marketing spend, its efficiency, and possibly the appeal of advertising revenue?

Luis von Ahn executive
#62

You're correct that during the past 2 years, I've warmed up to ads, in terms of both ad revenue and ad spend. A few points to make; regarding ad revenue, for the foreseeable future, we will stay a subscription business. Our subscription business is far bigger than ads.

Mark Mahaney analyst
#63

Yeah

Luis von Ahn executive
#64

That's going to keep going the same way for some time. That said, ads look like a really big opportunity to us. Our user base is large, our active user numbers are large, and nearly every app at our size or bigger earns far more from advertising than we do. So there's a major opening there, and that's why we've gotten much more sophisticated about it. A few years back at Duolingo, our whole ad revenue investment amounted to half an employee. One person spent half their time on ads. That's the level of investment we had back then. These days we have a team. It's far more professional now. They understand what they're doing. My expectation is that across the coming few quarters, you'll notice some gains in our ads business. Getting the hang of this requires time, but what I'm seeing is that we'll simply be able to deliver much higher quality ads inside our product that also bring in more money for us. We're working on it. This is an active work stream for us. So that's that. On the ad spend side, meaning marketing for us, we've also become much more sophisticated. For a period, I was hugely allergic to performance marketing. How I look at it today is that getting addicted to performance marketing isn't something we want, since that's never good. But it is a useful tool to have alongside your other marketing strategies. Our marketing team, which in my view is probably the best marketing team anywhere, has now begun adding performance marketing as an additional strategy. We're quite pleased with that. We're seeing the results show up in our top of funnel because of it.

Mark Mahaney analyst
#65

Alright. Thanks, Luis. And thanks, Gillian.

Luis von Ahn executive
#66

Yes. Appreciate it.

Gillian Munson executive
#67

Thanks, Mark. See you.

Luis von Ahn executive
#68

Hi, Ygal.

Ygal Arounian analyst
#69

Hi Luis. Hi Gillian. I appreciate you both fielding these. So, based on what I'm hearing on this call and the topics being discussed, it seems like perhaps we're beginning to pivot, or at least place more emphasis on generating revenue—mentioning advertising and additional subscription options. Would you say that's accurate? I'm not suggesting you're abandoning the work on daily active user optimization, but it does appear that the revenue-generation aspect might be getting more attention. How are you approaching that kind of change? What's the appropriate moment for it? What does it entail? I've got another question after this.

Luis von Ahn executive
#70

Yes, you've picked up on that well. That's a solid question. Honestly, for a stretch of this year, one of our biggest areas of focus has been growing the number of people actively using our product. Our thinking is simple: having more users improves just about everything — it means a much bigger business, and so on. We've set ourselves a bold target of reaching 100 million daily active users. We're optimistic about hitting that number, and we believe we can get there. Once we reach it, every aspect of the business improves. With a larger user base, there's more money to be made — everything gets better. It's also true that when it comes to our experimentation, at the start of the year we eased off a bit on monetization. Basically, our attitude was, "Hold on, hold on — let's pause." If there's any way we're making money that works against growing DAU, we shouldn't pursue it. The truth is, several of our practices were hurting DAU growth. We went through something of a reset. Now that sufficient time has gone by, our grasp of the levers is much stronger; we've shifted back toward, "Alright, this kind of monetization works, that kind of monetization works." Take longer free trials as an example—we mentioned those. We get it now, and we're pushing hard on them. Advertising is another area where, honestly, if the ads we serve are much higher in quality and generate more revenue for us, that's a great outcome — it fits right in line with DAU as well. We're putting money behind it. It's fair to say we're likely treating it with more seriousness now. I'm not sure that's the best way to put it. Compared with half a year ago, we're giving it somewhat more attention. That part is accurate. I can't recall a moment when we've been like — many folks ask us, "So when will you quit focusing on DAUs and begin earning money instead?" Quite a few people pose that question to us. A change of that sort isn't something I expect from us. Our aim remains reaching 100 million DAUs in 2028, and we'll keep working toward it. Over time, you'll likely notice us placing more emphasis on monetization bit by bit, once we identify approaches to earning revenue that don't conflict with expanding DAUs.

Ygal Arounian analyst
#71

All right, that's quite useful. I'd like to revisit what you said regarding open source models and the way they've pushed AI compute expenses lower. First, how ought we view that chance when it comes to the margin structure as time goes on? It seems as though there are several levers available there. This is a compelling topic, since during this earnings cycle we've repeatedly heard the same thing from plenty of companies, this shift toward open source models. So then, in what way does product velocity shift? Since I'm aware AI has played a major role in fueling your product velocity. With costs declining, does that turn into a larger opportunity? Thank you.

Luis von Ahn executive
#72

Yes. Broadly speaking, it simply happens that, relative to twelve months ago, far superior open source choices exist today. Although in certain areas they trail the frontier labs, for the majority of tasks they aren't trailing by much, perhaps three to six months. It simply happens that plenty of applications don't require the very smartest model. The truth is that for numerous applications we rely on, the quality cannot be told apart. Take, for instance, handling a call, such as conversation practice with someone whose vocabulary in a language is only 100 words. There's no need for your model to match a philosopher. You'll likely discuss, I don't know, frying eggs. You won't be discussing anything highly sophisticated. We discover that often, we're able to move over to the open source model. Of course, internally we still rely on numerous models from, for instance, OpenAI, Anthropic, and that will keep being the case. It will simply always be a choice, a trade-off regarding whether we can achieve it at approximately equivalent quality, yet using the open source option. When that holds, we'll go with open source. There remain situations where we cannot, and in those situations, we'll keep on paying. My expectation is that what will occur is that the cost per usage of AI, whatever that may be, the cost per token, if you prefer that term, will decline for us, since we'll keep shifting increasingly toward open source. Also, as you mentioned, that likely means we can provide growing numbers of AI features to growing numbers of users. Overall, I'm quite enthusiastic about that.

Gillian Munson executive
#73

So the way I'd frame it is that when we began the year, we decided we'd be quite patient with ourselves regarding what the business model would ultimately look like. Our real effort has been to hold to the notion that bookings would increase 10%-12%, while revenue would rise 15%-18%. At the start of the year, an adjusted EBITDA margin of 25% seemed like the appropriate amount of investment measured against growth. What we're essentially telling you is that by moving up roughly a point and a half on adjusted EBITDA margin, we're finding AI-related cost savings that will hand us a somewhat structurally improved margin, and that holds true even though we aim to roll voice call out to every super user across the year.

Ygal Arounian analyst
#74

Very helpful. Thank you.

Luis von Ahn executive
#75

Thank you, Ygal.

Luis von Ahn executive
#76

Justin, hello.

Gillian Munson executive
#77

Hi, Justin.

Justin Patterson analyst
#78

Good afternoon, everyone. To start off, I'd like to understand what's underpinning your belief that things will speed up in Q4. Does it come down to favorable year-over-year comparisons, or is there genuinely something happening on the product front that's beginning to work? Luis, I'd also like to revisit the goal of keeping the mission within reach financially — specifically, making education accessible to a broad population. Clearly, when you push more advertising into that environment, a friction point emerges: users can become annoyed by the experience. Does that begin to trigger undesirable actions? I'd really like to understand more about how you're weighing the need to keep earning revenue throughout the company against the goal of making education broadly available here. Thanks.

Gillian Munson executive
#79

You're asking me to address the bookings guidance; regarding that guidance, the truth is that the way this year is unfolding hasn't shifted very much. Naturally, we've kept the guidance as is. Since Q2 came in somewhat above what we'd anticipated, what you're essentially seeing is us making an adjustment to reflect that. It's worth keeping in mind that roughly half a point of FX headwind is baked into our bookings guidance as well. When you add it all up, you're likely in a slightly improved position. From my perspective, things are unfolding in line with our expectations. We've consistently believed that the ramp-up would build gradually—this isn't the sort of environment where things move in a smooth, predictable line. We recognize that the initiatives we're pursuing need some time before their impact becomes visible, and that's exactly what's reflected in our decision to maintain the guidance, which suggests that Q4 will show improvement when measured by year-over-year growth rate.

Luis von Ahn executive
#80

Regarding your inquiry about keeping education within reach, this is an organization driven by its purpose. I hope that has come through. We'll stay a company steered by that purpose. Our aim is to create the finest education anywhere and deliver it to everyone, and that's what we're after. How we'll run things mirrors how we run them today, which is that our reach matters above all else. Each additional active user we bring in represents one more individual we're instructing, represents one more individual who, rather than mindlessly scrolling through social feeds, is genuinely picking up something of value to them. That's the focus we're placing. We also hold the view that it works out that our purpose — I genuinely believe that although we're an organization steered by its purpose, I think that if we achieve our aim, that also lines up with turning this into a very big business. The greater the number of users we gain, in some manner or another, the greater our ability to earn revenue from them.

Justin Patterson analyst
#81

Thank you.

Luis von Ahn executive
#82

Thanks, Justin.

Arvind Ramnani analyst
#83

Hey, Luis. Hey, Gillian. I hope things are going well for both of you. I've just got a few questions. With China increasingly turning into a significant geography for you folks, and given that you're both running OpenAI and Anthropic — or actually, I believe it's mainly OpenAI — to power your models, does that introduce any kind of data residency concern or regulatory exposure? There's simply a lot happening right now when it comes to regulation and protectionist-type policies. Or would you instead just shift toward a local model within that market? How are you approaching this? Because Luis, I'm aware you tend to think ahead quite a bit on AI rather than reacting after the fact. We'd just really like to hear — not that you're going to make any changes — how you're looking at it.

Luis von Ahn executive
#84

Yeah. That's right. I appreciate the question about China. For us, China is a really exciting market. It continues to expand. In terms of daily active users, it ranks as our second-biggest market, and we anticipate that within a year or two, it will be our biggest market on that same daily active user basis. On top of that, China monetizes quite well for us; the monetization there is roughly on par with France. It's a good, exciting market for us. As for AI, there's plenty to say. The truth is that within China, we just aren't able to use those AI models — the U.S. AI models, that is. Local models are what we must use. The law requires it.

Arvind Ramnani analyst
#85

All right.

Luis von Ahn executive
#86

That's what we do over in China. Every bit of our AI work relies on Chinese models. There's no other way to go about it. Regarding regulatory exposure, it's worth noting that we obviously recognize China carries a certain amount of regulatory danger. How that nation is run sits largely beyond anything we can influence. As for our government relations over there, we're feeling fairly confident; that part looks good to us. What the government might choose at some future moment, though, I can't say.

Arvind Ramnani analyst
#87

Great. And staying with that subject — could you provide a ballpark figure? I don't need anything exact, just an approximate sense of your AI spend. Is the bulk of it going to Anthropic and OpenAI, or do open-weight models play a part? And regarding cost relative to revenue, does AI account for 1%? Or is it 10%? Roughly speaking, how big are these AI costs for you?

Gillian Munson executive
#88

What we spend-

Gillian Munson executive
#89

Yes, what we spend on AI within cost of goods sold sits in the tens of millions of dollars, making it a meaningful portion of COGS. Hosting represents another substantial expense for us too, yet those are the two largest items flowing through cost of goods sold. Within the company itself, we likewise employ AI, and that figure lands nearer to the $10 million mark on an internal basis.

Arvind Ramnani analyst
#90

All right, that's ideal. Those seem to be the key items. Luis, should you have any wider perspective — right now there's considerable discussion regarding AI sovereignty, open-weight versus closed-loop models. If you could offer anything there, I'd appreciate hearing your big-picture take on it.

Luis von Ahn executive
#91

Broadly speaking, what we believe internally is that employing open-weight models wherever we can serves our interests best as an organization. Were I holding a magic wand, my effort would go toward shifting every last thing onto an open-weight model. That isn't always achievable, since the frontier models can be ahead at times. Looking at it from the company's perspective, it's simply much superior, given that open-weight models cost far less to use.

Arvind Ramnani analyst
#92

Got it. So would your expectation be similar to how things worked with hybrid cloud — do you figure that's the direction this lands? You'd have some form of orchestration directing a portion of traffic to the best-performing models and another portion to open-weight ones. Is that the way you see it developing?

Luis von Ahn executive
#93

My guess is that the majority of businesses leaning on AI in a big way keep a portfolio of models in play. In certain cases, you essentially have to go with one particular model, which happens to be proprietary. In other cases, you attempt to rely on open-weight models. My sense is that most businesses will maintain a portfolio like that — that's definitely the situation on our end. What I'd add is that we expect that within the coming period of time, that portfolio will tilt somewhat more heavily toward open-weight models compared to where it stands right now.

Arvind Ramnani analyst
#94

Perfect. Thank you so much.

Gillian Munson executive
#95

Thank you.

Luis von Ahn executive
#96

Much appreciated, operator. I want to express my gratitude to all of you for being here with us, and we're eager to connect with you again on our upcoming call.

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.