Duolingo Told Wall Street It Would Stop Squeezing Free Users. Nine Months In: Users +29%, Paying Users +14%.

In February, Luis von Ahn stood in front of shareholders and said the quiet part out loud: Duolingo would spend 2026 making the free product better, on purpose, knowing it would cost money.

"In 2026, we are deliberately prioritizing user growth and teaching better," he said in the February 26 results release, adding that the company would focus on the free learner experience "even though that moderates near-term financial growth."

That is a sentence almost no course creator, membership owner or community operator ever allows themselves to say. Deliberately monetize less. Watch the top line slow. Tell everyone in advance.

On November 3, the receipt arrived, filed with the SEC.

The trade, in numbers

Daily active users went from 50.5 million to 64.9 million, up 29%. Paid subscribers went from 11.5 million to 13.1 million, up 14%. Revenue rose 14% to $309.0 million and bookings rose 12% to $316.1 million.

That gap — 29% versus 14% — is the whole story. The audience grew more than twice as fast as the paying base. Duolingo poured roughly 14 million additional humans into the top of the funnel and converted a far smaller share of them than it used to.

The direction of travel across the year is even starker. In Q1, DAUs were up 21% and paid subscribers up 21%. By Q2, DAUs were up 23% and paid subscribers up 17%. By Q3, DAUs were up 29% and paid subscribers up 14%.

User growth accelerating. Conversion growth decelerating. Three quarters in a row. That is not an accident; that is a dial being turned.

The headline profit number looks apocalyptic and mostly isn't. Net income fell from $292.2 million to $27.2 million — but the year-ago quarter included a one-time $222.7 million income-tax benefit from releasing a valuation allowance. The cleaner comparison is adjusted EBITDA: $80.0 million last year, $78.7 million this year. Flat profit, 14 million more people.

The market did not punish it. The stock jumped 26% on the news.

The detail that changes the math

Here is the part worth stealing.

While the free tier was growing fastest, it was also getting cheaper to serve. Gross margin hit a record of roughly 73%, helped by lower AI inference costs. That is 72.9% against 72.5% a year earlier. CFO Gillian Munson told analysts the company had expected margin to be about two points lower this year; the gap came down to "better revenue and less AI costs than we expected."

The specific number is the one every creator building AI into a product should write down. "When we first launched Video Call, it was costing about $0.30 per call," von Ahn said of the company's AI conversation feature. It now runs at well under a cent per call, largely by moving to open-weight models Duolingo runs itself.

A 97% cost collapse inside roughly two years. Not a margin improvement. A category change in what you can afford to give away.

So they killed their own premium tier

Watch what Duolingo did with that savings, because it inverts standard pricing logic.

Video Call was the reason Max existed — the top tier priced above Super, effectively subsidizing expensive inference. Once the inference got cheap, the tier lost its job.

Munson confirmed the company has essentially stopped selling Max to new users and is moving Max individual subscribers to Super Family at a similar price. Unlimited Video Call is going to all Super subscribers by the end of 2026.

Von Ahn's reasoning on the call: the A/B test showed it was good for revenue, and it put Video Call in front of roughly ten times as many users, simply because Super has far more subscribers than Max did.

Read that again. They deleted the expensive tier, gave its flagship feature to the cheaper tier, and the test said revenue goes up.

And they are now planning to push AI features below the paywall entirely. Munson said the company will spend on inference for free users next year — "single-digit millions," in her words — because it wants free learners to have some of these AI features.

Why this is your problem too

Most people selling knowledge online run the opposite playbook. Gate harder. Shorten the trial. Strip the free tier until it hurts enough to buy. Every friction point is defended as "conversion optimization."

Von Ahn described the trap precisely on an earlier call this year: he argued Duolingo was simultaneously under-monetized, with low subscriber conversion compared with peers like Spotify, and over-monetized, with too much friction in the free tier. He also named the uncomfortable flip side — that a genuinely excellent free product makes paid acquisition harder, because users are too happy without a subscription.

That is the real tension in every freemium course business, stated by someone with 64.9 million daily data points.

The tactic they replaced friction with is almost boringly simple. The Q2 letter described concentrating monetization work on increasing revenue without adding friction for free users, with longer free trials delivering good early results and expected to lift both revenue and DAUs.

Longer trial. Less gating. More AI in the free experience. Slower subscriber growth now, bigger base later.

The next quarter is the actual test

The experiment is already entering phase two. In the Q3 release, von Ahn said the company would "keep making Duolingo even more fun and effective, while gradually putting more focus on growing bookings" in ways that preserve the quality of the free experience — the first public signal that the squeeze comes back, carefully.

Analysts noticed. William Blair's team wrote that the quarter begins to address the key worry after Q2 — whether accelerating engagement actually converts — noting that "DAU growth accelerated for a second consecutive quarter" and management expects paid-subscriber growth to bottom within a couple of quarters.

Duolingo guided Q4 to $322 million of revenue and $382 million of bookings. The stated medium-term target is 100 million daily active users.

If the thesis holds, the bigger, less-squeezed audience monetizes harder in 2027 than the smaller, harder-squeezed one would have. If it doesn't, Duolingo will have spent a year buying attention it couldn't bank.

Either way, somebody else paid for the test and filed the results publicly.