Duolingo — Blitz Index 56 (C)

Growth is decelerating — the lock-in isn't.

Last re-scored: October 15, 2026.

Score breakdown

Think we scored it wrong? Score it yourself →

Radar axes (0–100): Moat 62 · Speed 64 · Creative 60.

Watch item

Gates: G1 ✓ · G2 borderline (bar ≈ $11.45B vs $19.7B TAM). 9 points below the threshold — first score. Debt-free, $1.18B cash.

Strengths

• 60M DAU approaching
• FCF $360M FY2025
• DET scale

Concerns

• Bookings deceleration (−59% price in 12 months — price not scored)
• Permanent governance discount

Research notes

Q3 2026 update

Q3 2026 update (company release, Nov 3, 2026; quarter ended September 30, 2026): DAU 64.9M (+29% YoY, vs 50.5M); paid subscribers 13.1M (+14%, vs 11.5M); revenue $309.0M (+14% YoY, vs $271.7M) — subscription revenue $270.0M (+18%), subscription bookings $278.4M (+16%); total bookings $316.1M (+12% YoY); gross profit $225.1M (+14%), gross margin 72.9% (vs 72.5%); GAAP net income $27.2M, GAAP diluted EPS $0.54 (vs $5.95); adjusted EBITDA $78.7M (25.5% margin, vs $80.0M); free cash flow $75.3M (24.4% margin, vs $77.4M). One-time/non-operating GAAP item (labelled, not operating performance): Q3 2025 net income included a one-time $222.7M income-tax benefit from the release of a valuation allowance on deferred tax assets; Q3 2026 operating expenses included an $11.4M accrual for a company-wide performance bonus tied to a Q4 DAU growth goal of 25% or greater. Guidance from the same release (forward-looking): Q4 2026 revenue $322M (+14% YoY), bookings $382M (+13%), adjusted EBITDA $84M (26.1% margin); FY2026 revenue $1,221M (+18%), bookings $1,296M (+12%), adjusted EBITDA $323M (26.5% margin). Score 56/C unchanged — this is a quarterly update note, not a rescore. Source: Duolingo Q3 2026 shareholder letter, investors.duolingo.com (also furnished as 8-K Ex-99.2, Nov 3, 2026, https://www.sec.gov/Archives/edgar/data/1562088/000162828026069474/q3fy26duolingo9-30x26share.htm); press release filed as 8-K Ex-99.1, https://www.sec.gov/Archives/edgar/data/1562088/000162828026069474/q3fy26duolingo9-30x26xpres.htm

**Official score, BlitzLab v1.0 (frozen 2026-10-10), as of T = 2026-10-15:** Speed (A) 16/25 — A1 Growth 7, A2 Capital velocity 3, A3 Hiring/productivity 3, A4 Share velocity 3; Moat (B) 37/60 — B1 Network effects 8, B2 Distribution scalability 12, B3 Market size 8, B4 Gross margin 9; Creative destruction (C) 9/15 — C1 Order displaced 3, C2 Irreversible behaviour change 3, C3 Cost-structure change 3; Demerits (D) −6 — D1 Debt 0, D2 Revenue quality 0, D3 Governance −4, D4 Regulatory and social cost −2. **Total: 16 + 37 + 9 − 6 = 56, Grade C (40–59), 9 points below the 65-point threshold. No FLAG applies, because total demerits of 6 are below the 15-point FLAG threshold.** This is an assessment of the business against an academic blitzscaling framework, not a recommendation about any security. Figures below are drawn from the score file; where the score file labels a figure as reported, estimated, or unconfirmed, that label is preserved.

Thesis: a habit machine that has stopped sprinting and started compounding

Duolingo is the clearest example in this coverage of a company whose blitzscaling story has changed tense. The first act — grow at any cost, win the category, worry about the economics later — is largely over. What remains is something rarer and, in scoring terms, harder to reward: a very large daily habit, defended by gamification and by a content engine that artificial intelligence has made dramatically cheaper to run, inside a balance sheet that carries no debt. The score of 56/C reflects that transition faithfully. It is not a failing grade, and it is not a near miss caused by one weak item. It is the arithmetic of a company that scores respectably almost everywhere, excellently almost nowhere, and carries one structural deduction that does not change from quarter to quarter.

The central tension in the Duolingo story, as the score file presents it, is between durability and deceleration. On one side, the user base keeps growing, retention is at a reported record high, revenue has grown for twelve consecutive quarters, and the company is profitable on a GAAP operating basis with more than a billion dollars of cash and short-term investments. On the other side, the rate of growth is slowing on every measure that leads the business — quarterly revenue growth, bookings guidance, and the margin trajectory as AI costs enter the cost base. Duolingo, in the score file's phrase, is a company in the mature phase of its first blitz: less "grow fast" and more "stay, and stay profitably." The question the rest of this body tries to answer is whether that durability can become a second growth curve, or whether it simply becomes a very good, steadily maturing software business.

Duolingo was first scored in this programme without a blind process, as of 15 October 2026, using the latest reported quarter available at that date, the second quarter of 2026, reported on 5 August 2026 with its Form 10-Q filed on 6 August 2026. The company listed on NASDAQ on 28 July 2021 at an offering price of $102. Its third-quarter 2026 results, due on 3 November 2026, fall after the scoring date and are therefore excluded from the score; they are noted below only as a future marker, with no anticipated outcome attached.

Speed: still growing, but the slope is the story (A = 16/25)

Growth, item A1, is scored 7 out of 10, and the evidence explains both the generosity and the restraint of that mark. Revenue rose from $250.8 million in fiscal 2021 to $1,037.6 million in fiscal 2025, an increase of 38.7% in that final year on a confirmed SEC basis, and trailing-twelve-month revenue at the scoring date stood at $1,145.1 million on a confirmed calculation. The quarterly path tells the more important story: $271.7 million in the third quarter of 2025, up 41.1% year on year; $282.9 million in the fourth quarter of 2025; $292.0 million in the first quarter of 2026, up 26.6%; and $298.5 million in the second quarter of 2026, up 18.3%. Twelve consecutive quarters of revenue growth is a genuine achievement, and it is why the gate on sustained growth passes on confirmed evidence. But a sequence that moves from 41% growth to 18% growth in three quarters is deceleration, not acceleration, and the score refuses to reward the level of revenue while ignoring the direction of its growth rate.

The forward-looking measures point the same way. Fiscal 2026 guidance, as announced by the company and aggregated in reporting, calls for revenue growth of 16.3% and bookings growth of 10.9%, against bookings growth of 33% in fiscal 2025. Bookings matter here because, for a subscription business, they lead reported revenue: a subscriber who books today becomes revenue over the months that follow. A bookings guide that falls from 33% to about 11% is therefore an earlier and sharper signal than the revenue deceleration alone, and it is the single most important number in the speed section. The score file records the chief executive's own framing of 2026 as a year oriented more toward user growth and education quality than toward monetisation, which helps explain the deliberate quality of the slowdown without excusing it in scoring terms. A company can choose to invest in the product rather than harvest it; the model simply records that the growth rate, while the investment is being made, is lower.

User growth, on a press-reported basis, remains strong in absolute terms even as revenue growth slows. Daily active users reached 58.7 million in the second quarter of 2026, up 23% year on year, monthly active users reached 140.6 million, up 10%, and paid subscribers reached 12.7 million, up 17%. Daily active users have grown roughly fivefold since the initial public offering, and have risen for five consecutive years. Put those figures together and the implied paid conversion rate is roughly 9% of monthly active users — a large free audience with a minority paying, which is exactly the structure a freemium education product is designed to produce. The score treats these user figures as reported rather than confirmed, because they come from earnings releases as aggregated in reporting rather than from XBRL filings, and this body preserves that distinction.

Capital velocity, item A2, is scored 3 out of 5, and the reasoning is instructive about what this model does and does not reward. Duolingo has no borrowings in the sense recognised in its XBRL filings, a confirmed fact, and held $1,180.9 million in cash as of 30 June 2026, confirmed, plus roughly $1.3 billion in short-term investments. In February 2026 the company authorised a $400 million share repurchase programme with no expiry, and $70.3 million had been executed in the first half of 2026, on a confirmed and reported basis. Access to capital, in other words, is of the highest order. But A2 measures the speed of raising and deploying capital for blitzscaling, not the comfort of holding it, and a company in a capital-return phase — buying back its own shares rather than raising money to fund expansion — does not earn additional points for that return. The score is therefore a study in model discipline: the strongest possible balance sheet earns only a middling capital-velocity mark, because the capital is being distributed rather than weaponised for growth.

Hiring and productivity, item A3, is scored 3 out of 5 on a similar logic of restraint. The company employed more than 900 people according to its fiscal 2025 Form 10-K, a confirmed figure, and revenue per employee works out at roughly $1.15 million for fiscal 2025 and roughly $1.27 million on a trailing-twelve-month basis, both confirmed calculations. That is a software-grade level of productivity, and it supports the broader picture of a business whose economics scale without a proportional increase in headcount. What is missing is evidence about the pace of hiring itself: the score file records that the trend in headcount at quarter-end could not be verified, so the speed of team expansion cannot be demonstrated, and the item is scored conservatively at 3 rather than on the strength of the productivity ratio alone. A reader who wants to raise this item in a future quarter should look first for a verifiable hiring trend, not for another productivity ratio.

Share velocity, item A4, is scored 3 out of 5. Duolingo is, on a press-reported basis, the world's largest language-learning application and the leading application in the education category in many markets. Retention reached a reported record high of 84% in the second quarter of 2026. Those are category-leadership facts, and they are real. What is missing, and what caps the item, is a quantified market share measured on a spending basis: the score file records that such a figure could not be verified, so the full mark is held back. Category leadership in downloads and daily use is not the same as a measured share of education spending, and the model keeps the two distinct.

Moat: the habit is the moat, and distribution is its amplifier (B = 37/60)

If speed is the section where Duolingo's score is capped by deceleration, moat is the section where the score is earned. At 37 out of 60, it is the largest single contributor to the total, and it is built from two complementary strengths: a daily habit that is hard to break, and a distribution model that has become steadily cheaper as it has scaled.

Network effects, item B1, are scored 8 out of 18. That is a middling mark on the model's largest single item, and the score file is careful about why. Duolingo does not enjoy the kind of direct network effect in which each additional user makes the product more valuable to every other user in the way a marketplace or a communications network does. What it has instead is a data flywheel and a habit lock-in, and the score file weighs both. On the data side, content production measured in learning units rose from about 1,800 units per quarter in 2024 to 20,500 units in the first quarter of 2026, a more than tenfold increase on a press-reported basis drawn from shareholder letters as aggregated in reporting. Each additional learner generates data that improves the product, and the AI-driven content engine converts that improvement into new courses at a speed that would have been impossible under a human-authored model. On the habit side, the streak, the leaderboard, and the gamified loop produce the reported 84% retention rate, and the absolute scale — 140.6 million monthly active users on a reported basis — means that even a modest per-user effect aggregates into something substantial. The model's anchor for this band describes data and ecosystem effects that exist but are weak, and the score file places Duolingo at the top of that band, at 8, to reflect the quality of the habit lock-in. It is a fair placement: the flywheel is real, it is quantified in the content-production figures, but it is not the self-reinforcing marketplace effect that earns the highest marks in this item.

Distribution scalability, item B2, is scored 12 out of 18, the strongest moat item, and it rests on the best quantified evidence in the moat section. The core distribution model is mobile, direct-to-consumer, and organic: a free tier distributed through app stores, converted by the product itself rather than by a sales force. Alongside that core, the company has built business-to-business and business-to-government channels in the Duolingo English Test, accepted by more than 6,000 institutions across 110 countries on a reported basis, and in Duolingo for Business. The quantitative evidence for distribution leverage is the trend in sales and marketing expense as a share of revenue, calculated from XBRL filings on a confirmed basis: 23.6% in fiscal 2021 falling to 12.1% in fiscal 2025. A company that roughly halves its marketing intensity while more than quadrupling its revenue is demonstrating distribution leverage in the most literal sense — each incremental user costs less to acquire than the last. The model's anchor for this band describes a strong viral loop or a large distribution partner, and 12 sits squarely in that band. It is the item that most clearly separates Duolingo from an ordinary education company: the product is the distribution.

Market size, item B3, is scored 8 out of 12, and it is inseparable from the gate verdict on market size, which the score file records as a borderline pass. The arithmetic is worth setting out plainly, because it explains both the pass and the modest item score. The gate requires that the ten-year ceiling of the new market be at least ten times current revenue. Current revenue on a trailing-twelve-month basis is $1,145.1 million, so the bar is approximately $11.45 billion. The global language-learning platform market is estimated at $10.45 billion in 2025 rising to $19.69 billion in 2030, according to ResearchAndMarkets, an estimate cited via GlobeNewswire on 3 September 2026. The 2030 figure clears the bar, but only just: it is about 1.7 times the bar, which implies that capturing the whole platform market at a single point in time would still require an implausibly high share — the score file notes that the platform market alone would imply a share requirement of around 58% against the bar. The pass therefore rests on two legs: first, the digit test is met, because $19.69 billion exceeds $11.45 billion; and second, expansion beyond language learning, into mathematics, music, chess, and the English test, raises the ceiling above the platform market alone. That second leg is an option, not yet a demonstrated scale, and the item score of 8 reflects exactly that: a market large enough to pass, but with thin headroom unless the broader education expansion proves itself. Readers should treat every market-size figure in this section as an estimate, not a confirmed fact.

Gross margin, item B4, is scored 9 out of 12, and it is the most stable part of the whole score. GAAP gross margin was 72.4% in fiscal 2021, 72.2% in fiscal 2025, and 72.6% in the second quarter of 2026, all confirmed. GAAP operating margin moved from negative 23.9% in fiscal 2021 to positive 13.1% in fiscal 2025, also confirmed — a completed turnaround from loss-making growth to profitable scale. Adjusted EBITDA margin was 29.5% in fiscal 2025 and 25.9% in the second quarter of 2026, a modest compression that the score file attributes to AI costs entering the cost base. Fiscal 2026 gross-margin guidance of around 69%, as announced by the company, points to further compression ahead. The item earns 9 rather than the full mark because the margin, while high and stable over five years, is now compressing at the margin rather than expanding. That is the correct direction of travel to record: a 72% gross margin business that guides to 69% is still a high-margin business, but it is no longer a margin-expansion story.

Taken together, the moat section describes a company whose defensibility comes from behaviour and distribution rather than from technology exclusivity or from market size alone. No single moat item earns a top-band mark, and that is the point: Duolingo's moat is broad rather than deep, a combination of habit, data, organic distribution, and margin that is individually good and collectively strong, but that lacks the winner-take-all network effect or the enormous uncontested market that would push the section toward the top of its range.

Creative destruction: displacing the classroom, then displacing its own content factory (C = 9/15)

Creative destruction, at 9 out of 15, is scored evenly across its three items at 3 each, and the evenness is itself informative. Duolingo has already completed one act of displacement and is in the middle of a second, narrower one, and neither act is yet the civilisation-scale transformation that earns the highest marks.

The first displacement, item C1, scored 3 out of 5, is the replacement of the traditional language classroom and textbook by a free, app-based, subscription-supported model. That displacement is, on a confirmed qualitative basis, already accomplished: Duolingo is the category leader, and the old order of scheduled classes and printed materials is no longer the default way a new learner begins. The second displacement now under way is the replacement of the human-authored content production order by an AI-first production order. The score file records the chief executive's memorandum of 28 April 2026, on a press-reported basis, announcing the gradual discontinuation of contractor work as AI-generated content takes over. That is a genuine change in how the product is made, and the tenfold increase in quarterly content production is its quantitative footprint. But its scope is so far limited to the production process rather than to the structure of education itself, and the item is scored at 3 to reflect a displacement in progress with a bounded range.

Irreversible behaviour change, item C2, is scored 3 out of 5. The evidence is the daily learning habit itself: 58.7 million daily active users on a reported basis, with retention at a reported record high of 84%, and an internal medium-term goal, stated by the chief executive, of 100 million daily active users. A daily habit practised by tens of millions of people is a real behavioural shift, and it is more durable than a habit practised weekly or monthly. But the score file draws a deliberate distinction between a habit that is entrenched at the level of the individual and the firm, and a change that is irreversible at the level of civilisation — the kind of change in which the old behaviour becomes unthinkable rather than merely unfashionable. Daily language practice on a phone is the former, not yet the latter, and the item is scored accordingly.

Cost-structure change, item C3, is scored 3 out of 5, and it is the most directly quantified of the three. The collapse in content-production cost is visible in the production figures already cited: about 1,800 learning units per quarter in 2024 against 20,500 in the first quarter of 2026, on a press-reported basis. The chief executive's statement that generative AI features are profitable, set against a third-quarter 2025 gross margin of 72.5% on a reported basis, is consistent with margin defence rather than margin collapse during the AI transition. What prevents a higher mark is the absence of a published, quantified unit-cost measurement: the score file records that no direct per-unit cost figure has been disclosed, so the cost collapse is demonstrated by output volume rather than by a measured cost per unit. Volume is strong evidence, but it is not the same as a cost curve, and the conservative mark preserves that distinction.

Deductions and risks: a clean balance sheet, a concentrated vote, and a lawsuit that is not yet decided (D = −6)

The deductions section is where Duolingo's score is most distinctive, because it combines the cleanest possible position on debt and accounting with the most concentrated governance structure in this coverage and a live legal overhang. The net deduction of 6 points is small, and no FLAG is triggered, but the composition matters more than the total.

Debt, item D1, is scored 0, meaning no deduction is triggered. The company has no borrowings recognised in its XBRL filings, a confirmed fact, and held $1,180.9 million in cash as of 30 June 2026, confirmed, against equity of $1,409.7 million. There is no financial-distress trigger to record. Revenue quality, item D2, is also scored 0, with no trigger. The auditor is Deloitte, which issued an unqualified opinion on the fiscal 2025 Form 10-K, confirmed, and there is no history of revenue restatement, confirmed. One qualification must be stated plainly, because it affects how the profit figures in the speed and moat sections should be read: of the $292.2 million of net income reported in the third quarter of 2025, $256.7 million was a one-time tax benefit from the release of a valuation allowance, confirmed in the notes to the Form 10-K. That one-time item represents about 62% of fiscal 2025 net income of $414.1 million. The score file is explicit that this is not an accounting irregularity, and that the speed and moat assessments separate recurring profit from this one-time benefit. This body follows the same separation: the operating-margin turnaround and the gross-margin stability are recurring-economics facts, while the headline net-income figures for the affected periods should not be read as recurring profit.

Governance, item D3, is scored −4 out of a possible −5, and it is the largest single deduction. Duolingo has a dual-class share structure in which Class B shares carry 20 votes each, confirmed in its definitive proxy statement filed on 17 April 2026. The two founders together control approximately 76% of the voting power: Luis von Ahn holds 40.0% and Severin Hacker holds 35.8%, both confirmed. The practical consequence, stated plainly in the score file, is that outside shareholders have effectively no influence over the composition of the board. Founder share sales of approximately $42 million in September and October 2026, on a press-reported basis and conducted under a Rule 10b5-1 trading plan, are noted as planned sales whose scale is nonetheless worth watching. The deduction is structural rather than episodic: because the voting concentration is a constant of the capital structure since listing, it does not vary from quarter to quarter, and there is no governance-improvement path visible within the current structure. The score file records this as a candidate for a future model anchor on founder supermajority voting power, and this body agrees that Duolingo is the natural anchor case.

Regulatory and social cost, item D4, is scored −2 out of a possible −5, a partial trigger reflecting litigation that is pending and unproven. A pension-fund securities class action was filed in approximately October 2026, on a press-reported basis, with a class period of 2 May 2025 to 26 February 2026, alleging misstatements about AI course quality; those allegations are unproven and the case is in progress. Investor investigations and lawsuit solicitations by firms including Faruqi, reported from April 2026 onward, are also noted on a reported basis. No government investigation has been reported. The partial rather than full deduction reflects that distinction: a filed, unproven allegation with no reported government action is a real overhang, but it is not an established finding, and this body makes no prediction about its outcome. The score file notes that the third-quarter 2026 results, due on 3 November 2026, will be the first earnings release after the filing; that timing is recorded here only as a calendar marker.

Two further risks belong in this section even though they do not appear as deductions, because they explain why a business with a clean balance sheet can still face a difficult period. The first is the paradox the chief executive articulated in the first quarter of 2026, on a reported basis, that Duolingo is simultaneously under-monetised and over-monetised: a vast free audience that is not yet fully converted, alongside a paying audience whose tolerance for further monetisation may be limited. That tension is the strategic form of the bookings deceleration — growth in users without a matching growth in bookings suggests that conversion, not reach, is now the constraint. The second is the gap between the quality of the business and the market's recent assessment of it, which the score file notes in its valuation note: bookings guidance was cut from 33% growth to about 11%, and the market repriced the shares sharply over the preceding twelve months, with a short interest of roughly 15% recorded from local FINRA data. This body records that context because the score file records it, and because the model's lesson from earlier work is that the business score deliberately does not look at price. No conclusion about valuation is drawn here, and no price outcome is anticipated.

What to watch: the second curve, the bookings line, and the court file

The score file's direction-of-travel notes point to three things worth watching, and each maps to a different part of the score. The first is whether the second growth curve becomes measurable. Expansion into mathematics, music, and chess, with 148 AI-generated new courses recorded in connection with that expansion in April 2025, and the English test's reach across more than 6,000 institutions and 110 countries on a reported basis, are the options that justify the borderline market-size gate. What would convert them from options into evidence is disclosure of their scale: revenue contribution, learner numbers, or institutional growth for the English test in particular, whose contribution the score file notes has not been disclosed on a sustained basis. Until such disclosure appears, the broader-education thesis remains a plausible ceiling-raiser rather than a demonstrated market, and the B3 item should be read in that light.

The second is the bookings line, which is the earliest indicator of whether the deceleration is stabilising. The score file's upward triggers include a bookings rebound above the guided level and daily active users passing 60 million; its downward triggers include a further bookings downgrade and an expansion of founder sales under the trading plan. This body adds no new trigger and anticipates no outcome for any of them. It notes only that, on the score file's own logic, a return to the 65-point threshold would require either a re-acceleration of growth in A1 — quarterly revenue growth back above 25% is the illustration the score file offers — or a demonstrated expansion of the addressable market in G2 through the broader-education thesis. Neither is a single-quarter event, and the chief executive's framing of 2026 as an investment year suggests that neither should be expected immediately.

The third is the legal file. The securities class action is unproven and in progress, and its dismissal, settlement, or continued litigation would each bear on the D4 item at a future scoring date. No outcome is anticipated here. What can be said now is that the D4 deduction is the only deduction in the score that could plausibly move in either direction on a litigation development alone, whereas the D3 deduction will not move at all without a change to the capital structure itself. That asymmetry is worth holding in mind: one risk is episodic and could clear, the other is permanent under the current structure and must simply be lived with.

Duolingo at 56/C is, in the end, a portrait of a company that has graduated from blitzscaling without failing at it. The habit is real, the distribution is genuinely leveraged, the margins are high and the balance sheet is unlevered, and the growth that built all of that is slowing. The score does not punish the company for maturing; it simply stops rewarding it for speed it no longer demonstrates, and it records, without flinching, that the people who built the company retain roughly three-quarters of its votes. A future score above 65 would need a new chapter — a second curve with disclosed scale, or a re-acceleration that the current guidance does not promise — and the evidence for that chapter, if it comes, will appear first in bookings, then in daily users, and only later in revenue. Until then, the honest summary is the score itself: 56, Grade C, nine points short, with the strongest foundations in the coverage and the clearest speed ceiling.

More Platforms companies

Other Platforms names on the Blitz Index, highest score first: Alphabet (83/A), Meta (78/B), Airbnb (76/B).

Next earnings

Q4 2026 · Expected

Coverage

Duolingo, Inc. · Q3 2026 due Nov 3, 2026 (AMC) · Score 56 · Grade C

Embed this score

Copy the HTML below to show this score on your site. The badge links back to this page and updates automatically when the score changes.

<a href="https://blitzlab.app/companies/duol/"><img src="https://blitzlab.app/badge/duol.svg" alt="Duolingo Blitz Index 56/C — BlitzLab"></a> Copy HTML

Preview:

BlitzLab scores businesses against a scholar’s standard — not investment advice. No price targets, no buy/sell recommendations. Figures are from SEC filings (XBRL) as labeled in the source scorecard.