The business
Duolingo teaches languages (and lately maths, music and chess) through a free app funded by advertising and a paid tier without it. Super Duolingo, Max and Family subscriptions were $873.4m of the $1,037.6m of 2025 revenue; advertising, the Duolingo English Test and in-app purchases were the rest. The company reports monthly and daily active users as fourth-quarter averages and paid subscribers at the period end, so the three numbers describe different things: 133.1m people used the app in an average month of the last quarter, 52.7m on an average day, and 12.2m were paying on 31 December.
The economics are those of any subscription business paid in advance: cash arrives at the start of a term and is recognised through it, which is why bookings run ahead of revenue and deferred revenue sits at nearly half a year’s revenue on the balance sheet. The costs are the app stores’ fees and hosting inside cost of revenues, and engineering, marketing and administration below gross profit, with stock-based compensation inside all of them.
The question
Two things make 2025 hard to read from the headline figures. Net income of $414.1m includes a $256.7m tax benefit that will not recur, from the release of the valuation allowance the company had carried against its deferred tax assets; without it net income was $157.4m. And adjusted EBITDA of $305.9m, the figure the company leads with, adds back $148.6m of stock-based compensation and related expenses to a GAAP operating income of $135.6m. So the question is what the business earns for a holder of its shares once employees have been paid in shares at their cost and the tax line is a normal one, and whether five more years of subscriber growth and operating leverage would justify what a discounted cash flow at ordinary inputs says the forecast is worth.
Reported, derived, editorial: how to read the model
The model keeps the three kinds of number on three sheets. Reported holds only what the company filed, each row with its note or XBRL tag, and nothing on it is adjusted. Derived holds arithmetic on Reported and nothing else: margins, ratios, revenue per subscriber, and the rebuild of adjusted EBITDA and free cash flow from the GAAP lines they come from. Forecast holds the assumptions, three cases of them, and every row says it is a judgement of this page. The P&L, cash flow and valuation sheets are formulas over the three, and the Checks sheet proves the reported figures agree with each other, the rebuilds tie to the company’s own figures, and the forecast identities hold.
The distinction matters more than it sounds. A reader who finds a number they disagree with should be able to tell at once whether it is the company’s, the arithmetic’s or ours. Most published models mix the three on one sheet, and the argument that follows is about the wrong thing.
Subscribers times what each pays
Subscription revenue is not forecast as a growth rate. It is the average number of paid subscribers in the year times the subscription revenue each one generates. In 2025 the average was 10.85m (the mean of 9.5m and 12.2m) and the revenue per average subscriber $80.50, up 6.7% from $75.47 in 2024 on price, the shift toward the Max and Family tiers and currency. Bookings per subscriber were higher, at $91.82, because bookings are collected ahead of recognition.
The base case carries subscriber growth from 28% in 2025 down to 10% by 2030, with revenue per subscriber up 3% a year, which takes subscribers to about 25m and subscription revenue to about $2.2bn. The downside halves the growth and holds the price; the upside keeps growth above 20% for three years. The revenue bridge on this page shows how much of the five-year increase comes from more subscribers and how much from each paying more, which is the first thing to argue about.
| Line | FY2024 | FY2025 | Source |
|---|---|---|---|
| Paid subscribers at period end (m) | 9.5 | 12.2 | Reported |
| Average paid subscribers (m) | 8.05 | 10.85 | Derived |
| Subscription revenue ($m) | 607.5 | 873.4 | Reported |
| Revenue per average subscriber ($) | 75.47 | 80.50 | Derived |
| Subscription bookings per average subscriber ($) | 90.77 | 91.82 | Derived |
| Paid subscribers / MAUs | 8.1% | 9.2% | Derived |
The non-GAAP measures, rebuilt
Adjusted EBITDA is rebuilt on the Derived sheet from net income: take out interest income, add back taxes, depreciation and amortisation, stock-based compensation and related employer taxes, and acquisition costs and earn-outs. For 2025 that is $414.1m less $45.2m plus a $231.7m tax benefit reversed, plus $14.4m, $148.6m and $5.7m, which is $305.9m, the company’s figure to the thousand dollars. The same rebuild for 2024 gives $191.9m. The Checks sheet holds both.
What the rebuild shows is the size of the add-back. EBITDA after stock-based compensation, which is just operating income plus D&A, was $150.0m, a 14.5% margin; the adjusted figure is 29.5%. The $155.9m between them is almost entirely shares given to employees. Free cash flow of $360.4m, the company’s other non-GAAP figure, is operating cash flow less property purchases and capitalised software, and it too adds back the $137.4m of stock-based compensation that operating cash flow includes. Free cash flow after it is $223.0m, a 21.5% margin, which is the figure the valuation sheet discounts.
None of this says the company should not report adjusted EBITDA. It says the number has to be read with its reconciliation beside it, and that the definition of free cash flow changed in 2025 (it no longer adds back taxes paid on equity awards or acquisition costs), so the FY2023 figure once reported is not comparable and the model derives it from the cash flow statement instead.
What the forecast cases imply
In the base case revenue reaches about $2.5bn in 2030, growth slowing from 26% to 14%, and operating leverage takes the GAAP operating margin from 13% to 19% as the cost ratios fade toward their targets; adjusted EBITDA margin moves from 29.5% to about 31%. Free cash flow after stock-based compensation rises from $223m to about $590m. Cash accumulates to over $4bn because the model assumes no buybacks, dividends or acquisitions, which is an assumption to change rather than a prediction.
The downside is not a collapse: subscriber growth of 12% fading to 4% with a flat price and weaker cost control still roughly doubles free cash flow after SBC by 2030. What it changes is the value, because so much of a discounted cash flow at these inputs sits in the terminal value. The upside holds growth higher for longer and takes the margins further.
What the valuation sheet is, and is not
The Valuation sheet discounts free cash flow after stock-based compensation at 31 December 2025 at the rate on the Forecast sheet, adds a perpetuity-growth terminal value, adds the cash and deducts the operating leases (there is no financial debt), and divides by the diluted shares. The result is the value per share the selected forecast implies at those inputs, and the Sensitivity sheet shows how far it moves with the discount rate, terminal growth and an exit multiple used as a cross-check. Beside it, as a memo, is the value if stock-based compensation were added back instead: it is higher, and it is wrong for a holder, because the shares paid to employees come out of the same share count the value is divided by.
The page does not compare any of this with the share price, and nothing on it is a target, a fair value or a recommendation. It is a model of what the business would have to do for the inputs to hold, which is the thing a reader can argue with.
What to change first
Three inputs move the answer most. Subscriber growth in the first forecast year sets the base for every year after; the revenue per subscriber growth compounds against it; and the stock-based compensation ratio by 2030 decides how much of the operating leverage reaches a holder. The discount rate and terminal growth move the value per share more than any operating input, which is a reason to treat the value as a consequence rather than a conclusion. The controls above change the base case; the full workbook changes any of them in any case.