Duolingo, Inc. (DUOL) Q2 2026 Earnings Report, Transcript and Summary

Good evening, everyone, and welcome to Duolingo's Second Quarter Earnings Webcast. Today after market closed, we released this quarter's shareholder letter, a copy of which you can find on our IR website at investors.duolingo.com. On today's call, we have Luis von Ahn, our Co-Founder and CEO; and Gillian Munson, our CFO. They'll begin with prepared remarks before we open the call for questions. And please note, this call is being recorded. Before we begin, please note that we'll make forward-looking statements regarding future events and financial performance. These statements are subject to risks and uncertainties described in our SEC filings and are based on assumptions we believe to be reasonable as of today, and we undertake no obligation to update them. We'll also discuss both GAAP and non-GAAP financial measures. Reconciliations can be -- of the two can be found in our earnings materials, and we encourage you to review them when evaluating our performance. And now I will turn it over to Luis.

Thanks, Debbie, and thank you all for joining. Q2 was a strong quarter. DAUs grew 23% year-over-year, accelerating from Q1 and coming in slightly ahead of our expectations. And we're encouraged by what we're seeing so far in Q3. The vast majority of that growth came from the work we do every day through what we call The Green Machine. We test hundreds of product changes, measure their impact and double down on what works. Most changes are small, but they compound over time. As I discussed in our shareholder letter, that's also what's driving CURR, a measure of user retention to an all-time high. Another highlight of the quarter was Streak Revival. A onetime campaign we ran in June. The idea was simple: give learners who lost their longer streak, a chance to earn it back by completing 3 lessons. More than 15 million learners revived their streaks. And what's particularly encouraging is that these users are also showing better retention than a typical reengaged cohort. We brought learners back to a product that keeps getting better at teaching languages, chess, Math and Music, and they're staying. We're still early in executing our strategy of prioritizing user growth and teaching better, but Q2 gave us more confidence that we're on the right track. With that, I'll turn it over to Gillian.

Thanks, Luis. Welcome, everyone. As Luis said, Q2 was a strong quarter. In addition to the DAU acceleration Luis just mentioned, topline results were in line with our expectations and profitability was slightly ahead of our plan. As we look at the remainder of 2026, I want to reiterate how we're managing the business. We are investing deliberately in the opportunities that we believe can make Duolingo a significantly larger business over the long term. At the same time, our team continues to operate with discipline. We continue to execute to our full year bookings and revenue target ranges of 10% to 12% bookings growth and 15% to 18% revenue growth. We have increased our target adjusted EBITDA outlook to 26.5% from the 25% we outlined at the start of the year. As for point estimates to help you build your models, please keep in mind the following: for the full year, we expect bookings growth of approximately 11% and revenue growth of roughly 16%. At constant foreign exchange rates, from our last call, the bookings growth rate would be about 0.5 point higher. For gross margin, we now expect to end the year closer to 70% as compared to the 69% we initially expected as we drive more AI content into our products offset by AI cost savings. We expect adjusted EBITDA to be approximately $320 million at the margin of roughly 25.5% I just mentioned. And we expect to generate over $375 million of free cash flow this year. While it is not included in our 2026 guidance, we do want you to know that the company has a bonus plan that will trigger if Q4 DAU growth is 25% or higher and would be paid out during Q1. Since it's currently uncertain whether that threshold will be met, we have not included it in our 2026 guidance. If it were achieved, we would expect the payout to be roughly $10 million in cash, potentially higher if DAU growth is higher. For Q3 itself, we expect bookings of approximately $307 million or growth of 9%. Revenue of $302 million, representing growth of 11%. We expect gross margin to be 71% and adjusted EBITDA of roughly $76 million, representing a margin of 25.2%. Our balance sheet and cash flow potential remains strong. We ended the quarter with $1.3 billion in cash and investments and generated $79 million in free cash flow. We repurchased about $44 million of stock during the quarter, bringing cumulative repurchases under our authorization to $72 million or approximately 700,000 shares. Putting it all together, our user momentum is strong, our business model continues to generate significant cash flow, and our team is executing well in an important investment year. We remain focused on reaching 100 million DAUs in 2028, and we believe the path there can create a significantly more valuable business for our shareholders. And now I'll turn it back to the operator, and we're happy to take your questions.

Your first question comes from Wyatt Swanson with D.A. Davidson.

Given DAUs are now expected to be above that 20% year-over-year growth in the second half of the year, could you maybe give some color as to why the full year bookings guide wasn't raised more? Like I realize you're in experimentation mode, but shouldn't a larger amount of users on the platform technically translate to increased bookings even if you're not pushing for monetization.

Yes, I think that that's a great question. I mean the first thing to note is that our users don't monetize immediately. I mean, they -- some of them monetized -- it takes a while for them to monetize because we have this premium model. We do expect that higher DAUs will imply higher revenue. But it's going to take some time. And the second reason is exactly what you said. We said at the beginning of this year that we were going to operate in this box in terms of revenue, and we are going to continue with this box. Basically, roughly 11% year-over-year bookings growth and then the rest of the efforts are in increasing DAUs and in teaching better.

Got it. Okay. That's helpful. And then you noted with the extension of free trials that helps improve both engagement and monetization. Could you just talk to some of the underlying mechanics as to how exactly that works going from 1 month to 2 months?

Yes. So the -- we're -- just to put it into context, our monetization team, their goal this year is to find things that monetize that are not at odds with user growth. I mean, historically, some of the ways that we have monetized have been by adding friction to the free use of product and that is at odds with user growth. So one of the things that has worked the best is longer free trials. And in particular, the main one that we're trying is historically Duolingo -- free trial in Duolingo and we say, "Hey, you can try for free. It's been 7 days. So we give you 7 days for free. That's what historically has happened. We are now shifting most of our free trials, not quite all of them yet to a 1-month free trial. And what -- the way that works is basically significantly more people decide to take that trial because it's just a better deal. And because of that, we got a larger number of people actually converting to payers. And the other nice thing is that as soon as they agree to go on the free trial, the experience just gets better because we turn off energy, we turn off the ads. So this actually increases daily active users as well. So we like it very much, and it's having a lot of good traction.

Your next question comes from the line of Andrew Boone with Citizens.

I wanted to ask about just your role in terms of influencers, how do we think about international marketing and kind of the changes that you guys highlighted in the letter? And then secondly, as we think about the U.S., Luis, understood the strength in the quarter and kind of resurrected users. Can you talk about top of funnel though? How do you feel about trends there and kind of the broader opportunity of attracting new users that may be new to Duolingo?

Yes. So I should say our marketing team, just to put it in historicals, most of our growth has been organic through word of mouth. In fact, for the first several years, it was 100% organic to word of mouth. Then we added mainly one form of marketing, which was social -- our own social media accounts. And that's still going really well. We are -- I mean we're getting more than 1 billion impressions per quarter on our own social media accounts, which is incredible. But our marketing team is really starting to expand to having other important tools. One of them is creators like groups of creators. And that's working quite well, actually, specifically in certain countries, countries like China, Indonesia and India. Something like 2/3 of our total social media impressions come from influencers. And what we're finding is that we are able to not spend a ton of money on this and what's nice is that these people have different audiences than us. And so this really just brings in a lot of new users. And this is something that we're very happy with. We use influencers differently in different countries, some countries more than others. But that's working really well. I should also mention that our marketing team has just become significantly more sophisticated on performance marketing. So while still the majority of our growth is organic, we're getting good traction on both usage of influencers and performance marketing. And we are seeing that in top of funnel. So it is increasing. So we actually feel pretty good about our top of funnel. It is increasing. And you asked about the U.S., in particular, the U.S. growth has increased quite a bit in the last quarter, and that's something that we're happy with. By the way, if you look at our growth, our DAU growth is really broad-based. Basically, all regions are growing and their -- all regions are growing faster than they were before. Asia is still the fastest growing but U.S. has increased.

One thing I might add to that, Andrew, is the top of the funnel is part of the story of the quarter. It got better. As you know, that's been a focus area of ours. We've talked about how we want to get better there. And in almost every region, the rate of growth there improved in top of the funnel in the quarter, which is a great accomplishment for the team.

Your next question comes from the line of Nathan Feather.

My end, both of them on the monetization angle. I guess, Interested to hear with the learnings you've seen so far from giving video call to new Super subscribers? And how is that influencing your plans for video call and Max generally as we go from here?

So video call is an interesting thing. We're -- I mean, we love this feature. It's an excellent feature for learning conversations, practicing conversation. It really works. We have, in fact, research that shows that if you use video call, you get better at conversation. So it really works, it's really good. When we first started adding video call to the platform, the first time that we put a video call in the platform, I remember the team that was working on it told me, okay, we can give this to users, but it's going to cost like $0.30 per call. to give to users. And that was expensive. And this is why we decided to put it behind our most expensive plan, which is Max but we said back then is if we can decrease the cost of this, we're going to try it in different places because it is our aim to give video call to as many people as possible because it really helps in learning a language. The good news is that through a lot of really hard work, we've been able to bring down the cost of video call. It is now under $0.01 per video call. And the reason for that is mainly a move towards open source models. It's just a lot cheaper to do that, and we don't see a loss in quality in there. So we're very happy with that. And because of that, we're able to now give video call to Super subscribers. The state that we're in right now is most new Super subscribers get video calls. So if you buy Super right now, you will get video call as well. We expect that over the next few months, we're going to give video call to existing user subscribers. We haven't quite done that to all of them. We expect that we're going to do that as well. And that's really good because more people are going to be able to practice conversation. Now that calls into question what are we doing with Max? And my answer is, I don't know yet. We're -- there's a few possibilities. One possibility could be that, while Super subscribers get a limited version of video call, like limited number of video calls and Max subscribers get unlimited. That is a possibility. Another possibility truthfully is that we may actually sunset Max. But -- what I will tell you is that we're going to have an answer to this in the next couple of quarters. And in addition to that, we're going to try to do this without a loss of revenue. And that's partly why we're not going super fast here because we're trying to figure out how to do this without losing much revenue. But again, our intent is that we give it to as many users as possible.

Okay. Great. That's really helpful. And then one thing we've seen reports that you're testing in ad-supported tier. So I guess help us think through, do you see any opportunity for maybe a lower priced tier, something in between Super and the free model and from a user segmentation perspective, what are you really going after there?

Yes. We are testing that. It's called Super Lite. It is being tested. And now I really want to emphasize the word tested. We are -- I don't know what's going to happen with Super Lite in the end. It is cheaper than Super, it's about half the price, depending on the geography. It is ad supported, so you get ads. And in addition to that, basically, you don't get unlimited energy, you get twice as much energy. So that's the idea. At the moment, it's still a small fraction of our subscribers are in Super Lite. But part of the reason is that we're early in the testing. We're just not advertising it very well. And of course, the goal with a Lite plan like this would be, if we're pretty sure that you're not going to buy Super then we should try to sell you Super Lite. That's kind of the idea. And you're going to see us experiment with that over the next few months. I don't know what will end up happening, but it's something we're trying.

Your next question comes from the line of Bryan Smilek with JPMorgan.

Luis, good to see CURR at an all-time high. Just curious, could you share more color on just overall retention by cohort. Are these new free trial users that are engaging on a daily basis, exhibiting higher engagement trends? Just curious anything you can add there from a retention perspective going forward.

Yes. So like we mentioned in the letter, our retention, our user retention, pretty much all metrics of user retention are at an all-time high. The one we look at the most, this one called CURR, which is current user retention rate. We're very happy that it is an all-time high and that it has increased by about a percentage point in the last year because tiny changes to CURR end up implying pretty large changes in daily active users over time because it really compounds. This is pretty broad-based. Basically, you see it in every region, you see across all types of users. The reason for this is just that the product is stickier. I mean basically, we have added a number of things, and it's hard to pinpoint to a single one because again, every single -- we have a new version of the app every single week and every version of the app has approximately 350 changes. So it's hard to point to something that actually did it. But it's just generally our product is just stickier and we're very happy with that because that's kind of the best type of growth you can expect that is just pure, it's a better product.

Great. That's super helpful. And then also kind of building on Nathan's question around Max overall. More from the speaking angle as well, can you just share more color in terms of overall engagement with more intermediate and advanced learners, which would, in my view, likely start to take on more speaking practice within the app.

Yes, we're getting -- I mean, -- so there's two things to say about that. The first one is that if you have access to video call, the engagement is very good. And in fact, that feature has gotten significantly better. One of the main metrics that we have for this feature is a number of words spoken by DAU who has access to that. That graph is a beautiful graph because it is entirely up into the right over the last couple of years. It's just every month, it is a little better than the previous month in terms of us getting you to speak more. And ultimately, this will just translate in people learning better and being engaged with it. So there's video call. We're also in the free tier, making people speak more. Just there's more speaking exercises and there's also more ways to answer, exercises that -- before you had to tap on the phone. Now you can answer it with your voice. So we're pretty happy with that. And it's exactly what you said. This is much more important to advanced and intermediate users, and we see that as something that will over time really help with certainly monetization, but also word of mouth because if people are learning better, they're going to tell their friends, et cetera. So we're very proud of that.

Your next question comes from the line of Ryan MacDonald with Needham.

Maybe just to ask on the bonus incentive comment sort of at the end of the prepared remarks. Obviously, the team internally is operating towards trying to get to that 25% rate. How should we think about that translating to, let's call it, the pace of new experimentation between now and the end of the year and if that picks up, are there any sort of features or initiatives that you'd call out that you're most excited about? Or is this really about just letting the changes in the experiments you've already made continue to sort of mature within the market and sort of let them sort of produce results?

It's a bit of both. So certainly, our rate of experimentation will continue. I mean it's pretty high this year. So if you look at actually a number of experiments that we're putting up per week, that keeps growing over the last year that has significantly increased. In fact, it is -- has increased faster than our head count. So that means that per person, we're actually putting out more experiments every week. So we're -- you will see that throughout the rest of the year. There's a lot of experimentation that's going to happen. And generally, that's across all the things that we experiment, teaching better and the daily active user growth are where we're spending our focus, but you're also going to see experimentation and monetization, et cetera. And it's related to the bonus in some way, but mainly it's just -- this year, our goal as a company was -- we decided we are going to focus most of our efforts of our incredible experimentation machine to just make it so that we can continue growing daily active users because we think that if -- as we said many times, if we can get to a company that is 100 million daily active users, that's just -- that's just a much bigger business, and that's where we want to get to. But you will see experimentation. You'll see a lot as you usually see from us.

Excellent. And then maybe on Max. You obviously are sort of working on a lot of things with Max right now and then one being sort of Super -- or voice being rolled out more to Super. Is there any sort of, I would call it, new feature product development that's sort of maybe geared or aimed towards the Max here at this point that could potentially sort of continue to extend the life of that tier? Or is really everything more focused sort of at the lower tier experimentation right now?

So I won't rule that out. It may happen that we develop something that we end up putting behind Max. We're developing a lot of features, and there may be some that because of cost of them or something we end up putting them behind Max. But that is not the goal. The goal really is to try to give speaking -- particularly speaking, features which are the most expensive ones to provide. The goal is to provide it to as many users as possible because, again, we just believe that the more people have access to this, the more word of mouth, the more users we have and the larger this business becomes. So that's the goal. But -- so I guess, what I'll say is that's not -- it's not what we're trying to do. But maybe that in 2 months, I come back and say, "Hey, Max actually got a new feature." And it's not because I'm trying to be secretive to you. I just don't know what will end up happening in terms of cost of certain features that we're developing.

Your next question comes from the line of Shweta Khajuria with Wolfe Research.

First one is on attention span for users as attention span, I guess, decreases, how is the engagement and session time across the app trending? And I guess what -- how are you addressing that? And then the second is on Math and Music. Could you please talk about just the product road map as you see Math and Music developed through the year and how that could drive contribution to DAU growth.

Yes. Thank you. I think part of your question is something that we've said in the past, which is just the world, in general, attention span is going down. I mean if you look at how people use things like social media apps, they use them in like 10-second bursts. Whereas for Duolingo, our lesson on Duolingo is maybe 2 minutes. That is something that we need to address. And we are working on that, trying to make the minimum session length longer. We are -- we have not yet -- we're testing stuff there, but we have not yet released anything. Part of the issue there is that if you make the minimum session length shorter, that's good, but you've got to make sure that people actually also come back many other times in the day so that it compounds. And that's kind of what we're trying -- what we're experimenting with but it is something we're actively experimenting with. In terms of Math and Music, so we're pretty excited about both of those. I should say there's still -- certainly when compared to chess. They still -- even though they do have single-digit millions of DAUs, they're much smaller than chess. So we do expect growth from both of them, but the contribution to our overall -- I mean we have 60-some million daily active users, even a 50% growth here doesn't contribute all that much but we are -- we do expect that they're growing -- we expect them to grow. Probably the easiest one to talk about in terms of strategy is Math because we are -- we understand that pretty well or at least better than Music, I think. With Math, originally, when we launched Math, I thought -- and I was wrong, I thought that we could get the average person on the street to get addicted to learning math. I was wrong. I don't believe we can do that. But -- so now our strategy is actually to mainly teach math to the people who actually need to learn math, which are basically K through 12 students. Now we may not sell to schools. That's not our goal to sell to schools, but it's at least that the user -- our user base for Math is mainly under 18. And as soon as we started working on that, that clarified a lot of things, and I think we started making a lot more progress rather than trying to get your average 35-year-old person who hates math to suddenly love it. I would love to do that, but I just -- I've given up on that one. And in Music, it's a bit early. I mean, we are working a lot on Music. I'll have -- we'll have more to say about Music in maybe a quarter or 2.

Your next question comes from the line of Mark Mahaney with Evercore.

Your comment about video call functionality. I'm sorry, Luis, were you saying that you could make it available to all Duolingo users or all Super subscribers?

Our goal -- I would love to make it available to all Duolingo users. I don't think we can do that right now. My goal in the moment is to make it available to all Super subscribers.

Okay. And the gating factor to -- for it not to be available to all users is just price, right, or cost, I mean, and you bring it out by another 90% next year, and you could do it?

Yes. But there's one other thing that we do have to take into account. This is one of the main things that gets people to buy. So if suddenly all users have it, there's less incentive to buy and then we would have to find other reasons to get people to buy. So there's a little bit of a trade-off here. So it's both cost and this is one of the main things that get people to buy.

Okay. And then I want to switch gears and ask you about advertising. And I think you've -- you've maybe changed a little bit over the last couple of years in terms of your view on both advertising revenue and on advertising spend. Where are you on that journey? Is there something that you've seen that's made you maybe a little bit more constructive on marketing spend and the efficiency of it and the desirability perhaps of advertising revenue?

You are right that over the last couple of years, I have come -- I've come around to add both in terms of ad revenue and ad spend. A couple of things to say. For the -- in terms of ad revenue, I mean, for the foreseeable future, we will remain a subscription business. I mean our subscription business is much larger than ads, and it will continue being like that for a while. However, we see a pretty large opportunity with ads. I mean we are -- we have a lot of users. We have a lot of active users and pretty much every app our size or larger than us, makes a ton more money from ads than we do. So we see a pretty big opportunity there, which is why we have become a lot more sophisticated on that over the last -- there was a time at Duolingo a couple of years ago, where our entire investment in ad revenue was half a person, there's one person whose halftime job was to deal with ads. So this is how much we were investing in them. We now have a team it's significantly much more professional. They know what they're doing. I would expect that over the next few quarters, that there -- that you're going to see some improvements in our ads business. This takes some time to get the hang of it. But what I'm seeing is we're just going to be able to give significantly higher quality ads in our product that also make us more money. So we're working on that. This is active work stream that we're doing, so there is that. In terms of ad spend, like for marketing, for -- marketing for us, we're also significantly more sophisticated. I mean it was a while where I was massively allergic to performance marketing. The way I see it now is I don't think that we want to get addicted to performance marketing because that's never good. However, it is a good tool to have to complement your other marketing strategies. So our marketing team, which I really think is probably the best marketing team in the world has now started adding as another strategy performance marketing and we're very happy with that. It's -- we're seeing the results in our top of funnel for that.

Your next question comes from the line of Ygal Arounian with Wedbush.

I guess, as I listened to the call and what we're talking about here, feels a little bit like we're starting to maybe shift or start to focus a little bit more on monetization, talking about ads and new tiers. Is that fair, not to say that you're moving away from the optimization around DAUs, but it does feel like you are maybe shifting a little bit the monetization side. How do you think about that transition? When is the right time? What's involved? And then I have a follow-up.

Yes. I mean it's perceptive of you. It's a good question. I mean the reality is, this year and for a while, we really -- one of the main thing that we're concentrating on is expanding our active user base because we just think that more users is just better for everything. It's significantly larger business, et cetera. And we have this aggressive goal to try to get to 100 million daily active users. We feel good about that goal. We think we can make it. And if we do get there, kind of all things get better, just more users, you can make more money, all things get better. It is true that our experimentation, when we started the year, we had a little bit of a pause in monetization in the sense that we're like whoa, whoa, whoa, wait, any monetization that we're doing that is counter to DAU growth, don't do it. And the reality is we were doing a number of things that were countered to DAU growth. So we kind of had a little bit of a reset. At this point, enough time has passed that we understand levers significantly better. So we are a little more back into like, okay, this type of monetization is good. This type of monetization is good. So for example, we mentioned longer free trials. We understand that, and we are kind of going hard on those. Ads is another business that we're perfectly happy if we have significantly better quality of ads and make more money from them, we're very happy with that because that's also DAU aligned. So we are funding that. So it is true that we're probably taking it more seriously, I don't know if that's the right term. We're focusing a little more on it than when compared to 6 months ago. So that is true. I don't think there's a time when we're like -- a lot of people ask us, "Oh, when are you going to stop worrying about DAUs and when are you going to start making money?" A lot of people ask us that. I don't think that we're going to have a shift like that. we're going to continue really trying to get to 100 million DAUs in 2028. But you'll see probably increased monetization focus kind of gradually over time as we find ways to monetize that are not at odds with DAU growth.

Okay. That's really helpful. And then I want to come back to the comments you made about the open source models and how that's driven down AI compute cost. So, a, how should we think about that opportunity in terms of the margin structure over time? It feels like there's multiple levers you could pull on that. And I mean, it's an interesting theme because we've heard this over and over again, this earnings cycle from a lot of companies in this -- like lean into open source models. And then how does that change the product velocity because I know AI has been a big component in driving your product velocity. So does that become a bigger opportunity with the cost coming down?

Yes. I mean, so at the highest level, it just turns out that there's -- when compared to a year ago. There's just significantly better open source options now. And while for some stuff, they are behind the Frontier labs, they're not that behind for most things, maybe 3 to 6 months. I mean it just turns out that for a lot of applications, you don't need the absolute smartest model, like the reality is the quality is indistinguishable for many applications that we use. For example, serving a call like a conversation practice to somebody who can only speak 100 words of a language. You don't need to serve a -- your model doesn't need to be as good as a philosopher. You're probably going to talk about I don't know, frying eggs, like you're not going to talk about very sophisticated stuff. So we find that in a lot of cases, we can switch to the open source model. Internally, of course, we're still using many models from, let's say, OpenAI and Anthropic, and we'll continue doing that. And it will just always be a decision -- a trade-off between, can we do it at roughly the same quality but with the open source? And if that's the case, we'll use the open source. There are still cases where we're not able to and in those cases, we'll continue paying. But I do expect that what will happen is that the cost per usage of AI like whatever it is, the cost per token, if you want to call it, will come down for us because we will continue moving more and more to open source. And I also, just like you said, that will probably mean that we're able to offer more and more AI features to more users. And so I'm pretty excited about that in general.

Yes. And the way to think about it is, we set out the year to be very patient with ourselves about what the business model is going to be. So we really have been trying to stick to this idea that we're going to grow bookings 10% to 12%, revenue 15% to 18%. And when we started the year at 25% adjusted EBITDA margin felt like the right level of investment relative to growth. What we're basically saying is by going up about 1.5 points on that adjusted EBITDA margin, we are seeing AI cost savings that will give us a bit structurally a better margin even with our goal to put voice call out to all Super users over the course of the year.

Your next question comes from the line of Justin Patterson with KeyBanc.

First, could you talk through just what's driving the confidence level behind that acceleration in Q4? Is that simply easy comps? Or are you just seeing something with the product side finally starting to click there. And then Luis, I just wanted to go back to making the mission affordable or making education affordable for the masses. Obviously, as you increase advertising in there, you do start to have a tension -- tension point in there, people get frustrated with the user experience in it. Does that start to cause some bad behaviors. So I'd love to hear more how you're thinking through the guardrails of still monetizing across the business while still having education be widely accessible within here.

Let me cover the bookings guidance. So on the bookings guidance, the reality is how the year is playing out hasn't really changed too much. Of course, we've held the guidance. Q2 was a little better than we expected. So you're basically seeing us adjust for that in the guidance. I think it's important to remember that our bookings guidance also includes about 0.5 point of FX headwind. So net-net, you're probably at a little bit of a better place. But I think the year is playing out the way we thought. We've always thought that acceleration is going to come over time. This is not a slow and steady. That's not the way this place is. But we know that it's going to take a little time for the work we're doing to play through, and that's what you see in terms of us holding which -- the guidance, which implies Q4 gets better from a year-over-year rate of growth perspective.

Yes. And your question about making education affordable. I mean this is a very mission-driven company. Hopefully, we have shown that we will continue being a very mission-driven company. Our mission is to develop the best education in the world and make it universally available, and that is our goal. The way we're going to be operating is similar to what we're operating right now -- how we're operating right now, which is -- we are -- first and foremost, care about our reach. And that every extra active users that we reach means there's one other person that we're teaching, means there's one other person that instead of scrolling -- doom scrolling on social media is actually learning something that is valuable to them. So we're -- that's what we're emphasizing. But we also believe that turns out that our mission -- I really do believe that even though we're a very mission-driven company, I think that if we succeed in our mission, that's also aligned with just making this a very large business. The more users we have in one way or another, the more we'll be able to monetize them.

Your last question comes from the line of Arvind Ramnani with Truist.

Just a couple of questions. Just with China just continuing to become a kind of important geo for you all, and you all are running OpenAI and Anthropic to run your models or I think mostly OpenAI, does it create some sort of data residency risk, regulatory risk, there's just -- there's just a lot going on in terms of regulation and sort of protection measures and would you just kind of move to a local model in that market? Just how are you thinking about it? Because Luis, I know you're fairly forward thinking about AI and not [ as reactive ]. So we just love to see -- not the you're making any changes, but how are you thinking about it?

Yes. So you're right. Thank you for asking about China. I mean, China is a very exciting market for us. It keeps growing. It is our second largest market in terms of daily active users, and we expect it to become our largest market in terms of daily active users in a year or 2. Also China monetizes pretty well for us, monetizes about as well as France. So we're -- it's an exciting good market for us. When it comes to AI, there's a lot there. The reality is that in China, we simply cannot use the AI models, the kind of U.S. AI models. We have to use local models. That's by law. So we do that. In China, all of our usage of AI uses Chinese models, that is just that -- you have to do that. And in terms of regulatory risk, it's important to mention that, of course, we understand that China poses some regulatory risk. That's kind of outside of our control, how they operate the country. But we feel pretty good about what we're doing there in terms of our government relations. So I feel pretty good, but I don't know what the government will decide at any point.

Yes. And just on this topic, right? Can you just give us just a rough estimate, right? Like I'm not looking for anything precise but just a rough estimate of what your AI costs are? And is it like largely on Anthropic OpenAI? Or are they kind of open-weight models? And just in terms of like your cost of revenues like does AI cost you like 1%, is it 10%? Just directionally, how significant are your AI expenses?

Yes. Our expenses on AI and cost of goods sold are tens of millions of dollars. So they're significant to the cost of goods sold. Hosting is another big cost for us as well. But those are the two biggies going through the cost of goods sold. Inside the business, we also use AI, and that is more closer to the $10 million range internally.

Yes. Perfect. That sounds great. Yes, I think those are the main questions. I mean unless Luis, if you have any like broader thoughts on just -- there's a lot of debate right now in like AI sovereignty, open-weight models or closed-weight models. Are you able to share anything, I love to get your high-level view on that.

I mean at the highest level, our view inside the company is that it is in our best interest as a company to use open-weight models as much as possible. So if I had a magic wand, I would try to move everything to an open-weight model. It's not always possible because sometimes the frontier models are more advanced. But from a company standpoint, it is just significantly better because it's way cheaper to use open-weight models.

Perfect. Perfect. And so would you expect like this sort of having a hybrid cloud, where do you think that's where we end up? Like you have some sort of orchestration where you put some traffic towards like the highest-quality models versus open weight, is that how you expect?

That's probably. My guess is that most companies that use AI heavily have a portfolio of models that they use where for some uses, you kind of have to pick this one model that happens to be a proprietary model. But for other things, you try to use the open-weight models. So my sense is most companies will have a portfolio model like that. That's certainly what we have. What I would say is that our expectation is that over the next some time, that portfolio will be weighted a little more towards open-weight models than it is today.

I'm showing no further questions. This concludes the Q&A section of the call. I would now like to turn the call back to the host for closing remarks.

Thank you, operator. I'd just like to thank everyone for joining us, and we look forward to seeing you on the next call.