---
title: "DUOL (Trans): DAU +23%, Max plan may be discontinued"
type: "News"
locale: "en"
url: "https://longbridge.com/en/news/295028818.md"
datetime: "2026-08-05T23:16:58.000Z"
locales:
  - [zh-CN](https://longbridge.com/zh-CN/news/295028818.md)
  - [en](https://longbridge.com/en/news/295028818.md)
  - [zh-HK](https://longbridge.com/zh-HK/news/295028818.md)
---

# DUOL (Trans): DAU +23%, Max plan may be discontinued

**Dolphin Research's Trans of Duolingo FY26 Q2 earnings call**

**I. Key takeaways**

1) Shareholder returns: Repurchased ~$44 mn of stock this quarter. Cumulative buybacks under the authorization reached ~$72 mn, or ~0.7 mn shares.

2) Full-year guide: Bookings and revenue ranges unchanged, with a higher profitability target. Bookings: Full-year growth range maintained at 10%–12% (point estimate ~11%); at the fixed FX rate used on the prior call, growth would be ~50bps higher. Revenue: Full-year growth range maintained at 15%–18% (point estimate ~16%).

Adj. EBITDA margin: Target raised to 26.5% from 25% at the start of the year, implying Adj. EBITDA of roughly $320 mn. GPM: Now expected to land near 70% for the year vs. 69% previously, driven by more AI-powered content offset by AI cost savings. FCF: Expected to exceed $375 mn for the year.

3) Q3 guide: Bookings of ~$307 mn (+9% YoY); revenue of ~$302 mn (+11% YoY); GPM of 71%. Adj. EBITDA of ~$76 mn, implying a 25.2% margin.

4) Q2 financials and cash: Revenue in line; profitability slightly ahead of plan. Ended the quarter with $1.3 bn in cash and investments; generated ~$79 mn of FCF in the quarter.

5) Contingent spend not in guide: A bonus plan triggers if Q4 DAU growth reaches or exceeds 25%, with payout in Q1. Given uncertainty on attainment, it is excluded from the FY26 guide; if triggered, cash outlay is estimated at ~$10 mn and could be higher if DAU growth is stronger.

**II. Call details**

**2.1 Management remarks**

1) User growth. DAU grew 23% YoY in Q2, accelerating vs. Q1 and slightly ahead of expectations; trends so far in Q3 are also encouraging.

Most of the growth comes from the daily 'green machine' workflow: testing hundreds of product tweaks, measuring lift, and doubling down on what works. Most changes are small, but they compound over time.

2) Retention at a record high. KER (existing-user retention) reached an all-time high, a direct outcome of product compounding. This also validates the strategy to prioritize user scale and better instruction quality.

3) Streak revival campaign. In Jun, Duolingo ran a one-off event allowing learners who lost their longest streak to reclaim it by completing three lessons. Over 15 mn learners revived their streaks.

This cohort retains better than a typical re-engagement cohort. They returned to a product that keeps improving across language, chess, math, and music, so they stayed.

4) Operating stance and long-term goals. Duolingo is deliberately investing in opportunities that can make the biz. significantly larger over time, while maintaining operating discipline; 2026 is positioned as an important investment year. Long-term goal unchanged: reach 100 mn DAU by 2028, a path management believes will create significantly more shareholder value.

**2.2 Q&A**

**Q: If 2H DAU growth is expected to exceed 20% YoY, why not raise the full-year bookings guide more? With more users on the platform, shouldn't bookings be higher even without pushing monetization? (D.A. Davidson)**

A: First, users do not monetize immediately. With a freemium model, some users convert after a lag; higher DAU should translate to higher revenue, but it takes time.

Second, as noted, we said at the start of the year we would operate within this revenue frame and we still plan to stay within it, i.e., bookings growth of roughly 11% YoY. The rest of the focus is on growing DAU and improving instruction.

**Q: Extending free trials is said to improve both engagement and monetization. What is the mechanism behind moving from one month to two months? (D.A. Davidson)**

A: For context, the monetization team’s goal this year is to find ways to monetize without conflicting with user growth. Historically, some monetization approaches added friction for free users, which conflicted with growth.

One of the best-performing tactics is a longer free trial. We historically offered 7 days; we are moving most (not all) free trials to one month. Because it is a better deal, more people opt in, and the absolute number that converts to paid increases.

Another benefit: once users enter a trial, the experience immediately improves—energy limits are disabled and ads are removed—so DAU also improves. We are very pleased with the results.

**Q: What role do influencers play in marketing, and how should we read the intl efforts and changes noted in the shareholder letter? (Citizens)**

A: Historically, most growth was organic, initially 100% word-of-mouth. We later added company-owned social accounts, which still perform well, generating over 1 bn impressions per quarter.

The team is now expanding tools, including creators, which work especially well in markets like China, Indonesia, and India. Roughly two-thirds of social impressions come from influencers, which are low-cost and reach different audiences, bringing in many new users; usage varies by country.

We have also gotten more sophisticated in performance marketing. While growth is still predominantly organic, both creators and performance marketing have made solid progress.

**Q: How is the top-of-funnel trending in the U.S., and what is the opportunity to attract new users? (Citizens)**

A: The top-of-funnel is expanding, and we feel good about it. U.S. growth picked up meaningfully last quarter.

Overall DAU growth is broad-based: virtually all regions are growing and at faster rates than before; Asia remains the fastest-growing, with the U.S. also re-accelerating. Improvement in the top-of-funnel is part of the story this quarter across nearly every region.

**Q: What have you learned after opening video calls to new Super subs, and how does this affect the roadmap for video calls and beyond? (Morgan Stanley)**

A: Video calls are excellent for conversation practice, and our research shows they improve speaking ability. Initially, the team estimated ~$0.30 per call—too expensive—so we put it in the highest Max tier, with the caveat that if costs came down, we would broaden access.

The good news: after significant work, cost per call is now under $0.01, largely by shifting to open-source models, with no observed quality loss. We can now offer video calls to Super subs: most new Super subs get access today, and we expect to extend access to existing Super subs over the coming months.

That raises the question of what to do with Max; no final answer yet. One option is limited video calls for Super and unlimited for Max; another is sunsetting Max. We will decide in the next 1–2 quarters and aim to avoid revenue loss, which is why we are not moving too fast.

**Q: Reports suggest you are testing an ad-supported tier. Do you see a low-priced tier between Super and free, and which users are you targeting? (Morgan Stanley)**

A: We are indeed testing a product called Super Light; emphasis on 'testing'—the end state is unknown. It is cheaper than Super, roughly half the price depending on market; it includes ads and offers energy that is not unlimited but roughly 2x the free tier.

Only a small subset of subs are on Super Light now, partly because tests are early and we have not promoted it. If we are confident a user will not buy Super, we may try to sell Super Light; we will experiment in the coming months, with outcomes TBD.

**Q: With KER at a record high, can you break out cohort retention? Do new free-trial users show higher daily engagement? (JP Morgan)**

A: As noted in the letter, nearly all retention metrics are at record highs. KER, our key focus, improved by roughly 100bps over the past year—small moves in KER compound into meaningful DAU gains over time.

The improvement is broad-based across regions and user types, driven by a stickier product. We ship a new app version weekly with ~350 changes per release; it is hard to attribute to any one change, but overall stickiness is up, which is the best kind of growth.

**Q: For speaking, how is engagement among intermediate/advanced learners? (JP Morgan)**

A: Two points. First, users with video-call access show very strong engagement, and the feature has improved notably; a key KPI—'words spoken per eligible DAU'—has trended up for two years, driving better learning and deeper engagement.

Second, beyond video calls, free-tier speaking is improving: more speaking exercises, and questions that were tap-to-answer on mobile can now be answered by voice. This matters more for intermediate/advanced users and should aid both monetization and word-of-mouth over time.

**Q: Regarding the Q4 DAU ≥25% bonus trigger, how will this influence the cadence of new experiments from now to year-end? If cadence accelerates, which features matter most, versus letting launched changes mature? (Needham)**

A: Both. Experiment cadence will remain high; weekly experiment volume has grown steadily over the past year, outpacing headcount growth, and will remain heavy through year-end.

Focus areas are 'better instruction' and DAU growth, with some monetization experiments as well. The bonus is a factor, but the primary goal is to devote most of the machine to sustained DAU expansion; reaching 100 mn DAU would make for a much larger biz., which is where we want to go.

**Q: For Max, are there tier-specific features in development to extend its lifecycle, or is effort focused on lower tiers? (Needham)**

A: Not ruled out. We are developing many features; some may end up in Max due to cost, but that is not the goal.

The real goal is to bring speaking functionality—the highest-cost set of features—to as many users as possible, which should boost word-of-mouth, users, and the biz. It is possible that in two months Max gets a new feature; that would reflect realized cost structures rather than a deliberate tier strategy.

**Q: As user attention spans shrink, how are in-app engagement and session length trending, and how will you respond? (Wolfe Research)**

A: Globally, attention spans are declining. Social apps are consumed in 10-second fragments, while a Duolingo lesson is ~2 minutes.

We are working to address this by shrinking the minimum viable session length. This is in testing and not yet launched; the challenge is to shorten sessions while ensuring users return multiple times per day to compound learning.

**Q: What is the roadmap for Math and Music, and how much can they add to DAU this year? (Wolfe Research)**

A: Both are exciting, but still much smaller than Chess—each has several million DAU. With ~60+ mn total DAU, even 50% growth in these categories will not move the overall needle much, though we expect continued growth.

Math strategy is clearer: initial hopes to hook the general public on math were wrong. We now focus on those who need math—K-12 students; we may not sell to schools, but the user base is primarily <18, which clarifies priorities and speeds progress.

Music is earlier; we are investing, and should have more to share in 1–2 quarters.

**Q: For video calls, will access extend to all Duolingo users or just all Super subs? (Evercore)**

A: We would love to offer it to everyone, but cannot today. The current goal is to offer it to all Super subs.

**Q: Is the constraint purely cost? If costs drop another 90% next year, would that enable universal access? (Evercore)**

A: There is also a monetization trade-off. Video calls are a major driver of paid conversion; if everyone gets it, purchase motivation weakens and we need other pay drivers, so we balance cost and monetization.

**Q: Views on ad revenue and ad spend seem to have evolved. Where are you now on performance marketing efficiency and the attractiveness of ad revenue? (Evercore)**

A: Yes, views have shifted over the past two years on both ad revenue and ad spend. On ad revenue, subscriptions will remain the core for the foreseeable future, much larger than ads for some time.

That said, the ad opportunity is sizable: with a large and active user base, comparable apps at our scale or larger make far more from ads. We have become much more professional here; two years ago, ad monetization was handled by 'half a person', now there is a team and we expect improvements over the next few quarters as we deliver higher-quality, higher-yielding ads.

On ad spend, we are also more mature. We still do not want to be addicted to performance marketing, but it is a good complement, and we are seeing results at the top of the funnel.

**Q: It sounds like you are leaning more into monetization—ads and new tiers. Is that right? How do you think about timing the shift while still optimizing for DAU? (Wedbush)**

A: This year and for some time, the primary focus remains expanding the active user base; more users make everything better and the biz. larger. 100 mn DAU by 2028 is aggressive but achievable, and reaching it improves everything.

We did pause monetization tests earlier this year when they conflicted with DAU growth and reset our approach. Now, with better understanding of the levers, we are leaning into those that align with DAU, like longer free trials and higher-quality ads, and we are willing to invest there.

So versus six months ago, yes, monetization has a bit more focus. But there will not be a hard switch away from DAU; we will push toward 100 mn DAU while gradually increasing monetization that does not conflict with growth.

**Q: Moving to open-source models lowered AI compute costs—what does that mean for margin structure long term? Does lower cost also expand the scope and speed of AI-driven product iteration? (Wedbush)**

A: Open-source options are much better than a year ago. While frontier labs still lead on some tasks, for many use cases the gap is only 3–6 months, and many applications do not need the absolute smartest model; in many of our scenarios, quality is indistinguishable.

So we can shift to open-source for many cases; we still use models from OpenAI, Anthropic, and others where needed. The rule of thumb: if open-source delivers roughly comparable quality, we use it; otherwise we pay for closed models. Over time, more open-source usage should reduce per-token costs and allow us to deliver more AI features to more users.

From finance: we intended to stay patient with the biz. model, hence the 10%–12% bookings and 15%–18% revenue frameworks. We set Adj. EBITDA margin at 25% early on as the right investment level vs. growth, and we now raise it by ~150bps, essentially saying AI-driven cost savings structurally support better margins—even as we aim to roll out voice/video calls to all Super users this year.

**Q: What drives confidence in a Q4 acceleration—base effects, or product impact finally flowing through? (KeyBanc)**

A: The full-year trajectory has not changed much; we maintained the guide. Q2 was slightly ahead, and the guide reflects that adjustment.

Note the bookings guide includes ~50bps of FX headwind, so on a net basis it may be a bit better. We have always expected acceleration to come through over time; maintaining the guide implicitly assumes better YoY growth in Q4.

**Q: As ads increase in the product, is there a risk of crossing a threshold that annoys users? How do you balance monetization with broad access to education? (KeyBanc)**

A: We are a mission-driven company, and we will stay that way. Our mission is to build the best education and make it universally accessible.

Operationally, we prioritize reach—every additional active user is another person learning something valuable rather than doomscrolling. We also believe mission and building a very large biz. are aligned: if we deliver on mission and grow users, we can find ways to monetize.

**Q: China is becoming a key market while you mainly run models from OpenAI/Anthropic—does this pose data residency or regulatory risks, or will you switch to local models there? (Truist)**

A: China is very exciting and still growing; it is our No. 2 DAU market today and could be No. 1 within 1–2 years. Monetization in China is solid, roughly comparable to France.

For AI in China, we cannot use U.S. models; we must use local models by law and we comply—AI usage in China runs on Chinese models. We recognize regulatory risks exist and outcomes are outside our control, but we feel good about our GR efforts; government decisions are inherently unpredictable.

**Q: Can you size AI costs? How much is Anthropic/OpenAI vs. open-source, and what share of COGS is AI—~1% or ~10%? (Truist)**

A: AI spend in COGS is in the tens of millions of dollars and is a significant component; hosting is the other major COGS line, and together they dominate COGS. Internally used AI spend is closer to ~$10 mn.

**Q: There is much debate over AI sovereignty and open vs. closed. What is your high-level view? (Truist)**

A: Our view is that it is in our best interest to use open-weight models wherever possible. If we had a magic wand, we would migrate everything to open-weight models; that is not always feasible, as frontier models can be ahead, but open-weight is much cheaper and thus preferable when quality is comparable.

**Q: Is the likely end state a hybrid orchestration—routing some traffic to the highest-quality closed models and the rest to open-weight? (Truist)**

A: Most likely yes. Heavy AI users will maintain model portfolios: some use cases require a specific closed model, while others can run on open-weight; we follow that approach and expect to tilt further toward open-weight over time.

<End of text\>

**Risk disclosure and statements:**[**Dolphin Research Disclaimer and General Disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

### Related Stocks

- [DUOL.US](https://longbridge.com/en/quote/DUOL.US.md)
- [MAX.US](https://longbridge.com/en/quote/MAX.US.md)

## Related News & Research

- [Duolingo Posts Double Beat in Q2: Here's Why the Stock is Falling Wednesday](https://longbridge.com/en/news/295024781.md)
- [Duolingo posts Q2 2026 results shareholder letter on investor relations site](https://longbridge.com/en/news/295008186.md)
- [Duolingo Q2 FY26 net income drops 26% to $33.16 million; revenue rises 18% to $298.45 million](https://longbridge.com/en/news/295088193.md)
- [MediaAlpha (NYSE:MAX) Director Sells $96,836.10 in Stock](https://longbridge.com/en/news/295030021.md)
- [What To Expect From QuinStreet’s (QNST) Q2 Earnings](https://longbridge.com/en/news/294903040.md)