Duolingo co-founder and CEO Luis von Ahn (@LuisvonAhn) said on the company's August 5th earnings call that Duolingo had reduced the cost of an AI-powered video call from roughly $0.30 to less than $0.01, giving the language-learning app room to distribute its conversation feature beyond its most expensive subscription.
The remarks, excerpted by The Transcript, explain the economics behind a product shift Duolingo has been testing throughout 2026: moving Video Call into the broader Super Duolingo subscription and deciding whether Duolingo Max still needs to exist.
Von Ahn has spent his career turning computing costs into mass-market products. Before founding Duolingo with Severin Hacker in 2011, the former Carnegie Mellon University computer science professor created reCAPTCHA, which Google acquired in 2009. Video Call applies the same distribution logic to generative AI. A feature built around real-time model inference becomes much more useful to Duolingo when its cost is low enough to reach millions of learners.
Open models changed the unit economics
When Duolingo first introduced Video Call, von Ahn said, the team estimated that each interaction would cost about $0.30. That expense helped determine the product's original placement inside Max, Duolingo's highest-priced subscription tier.
"It is now under $0.01 per video call," von Ahn said during the second-quarter earnings call. He attributed most of the decline to a shift toward open models, adding that Duolingo had not observed a corresponding drop in quality for the task.
The reduction is greater than 30-fold, based on the two figures supplied by Duolingo. The per-call measurement does not provide a complete view of Video Call's economics because costs can vary with conversation length, speech processing, animation, hosting and the model used. It still marks a material change in the marginal expense of serving the feature.
Von Ahn said open models can trail systems from frontier laboratories by three to six months in some areas, but that gap often has little practical effect on beginner conversation practice. A learner with a vocabulary of roughly 100 words does not require the most capable reasoning model available, he said. Duolingo continues to use models from OpenAI and Anthropic when their performance is required.
The company is moving toward a portfolio approach: proprietary models for tasks where they provide a meaningful quality advantage and cheaper open-weight models where users receive comparable results. Von Ahn said he expects Duolingo's model mix to shift further toward open weights.
The savings are large enough to affect Duolingo's financial outlook. Chief Financial Officer Gillian Munson said AI expenses within cost of revenue amount to tens of millions of dollars, while internal AI usage is closer to $10 million. Duolingo raised its full-year adjusted EBITDA margin forecast to 26.5% from the 25% outlook it provided at the start of 2026, with AI cost savings contributing to the improvement.
Max loses its defining feature
Lower inference costs have created a subscription problem for Duolingo. Video Call was one of the main reasons to pay for Max. Once conversation practice moves into Super, the top tier needs a new distinction.
Most new Super subscribers already had access to Video Call by the end of the second quarter, according to Duolingo's August 5th shareholder letter. Duolingo expects to extend the feature to existing Super subscribers later in 2026.
Von Ahn said Duolingo is considering several structures. Super could receive a limited number of calls while Max retains unlimited access. Duolingo could place another expensive AI feature inside Max. He also raised the possibility of retiring the tier.
That decision turns a technical cost reduction into a monetization test. Duolingo wants Video Call in the hands of more learners because better speaking practice can improve retention and word of mouth. The feature also gives users a reason to subscribe, so distributing it too widely could weaken paid conversion or reduce the appeal of a higher-priced plan.
Duolingo is trying to manage that trade while prioritizing user growth. The company reported 58.7 million daily active users in the second quarter, up 23% from a year earlier, and 12.7 million paid subscribers, up 17%. Revenue increased 18% to $298.5 million, while adjusted EBITDA declined 2% to $77.3 million as Duolingo invested in growth and broader product access.
The educational case rests largely on Duolingo's own research. The company's Efficacy Research Lab reported that learners using Video Call improved more on speaking assessments than comparison groups in studies covering beginner and intermediate learners. Those findings support von Ahn's push to widen access, though they were produced by Duolingo rather than an independent evaluator.
Open models have given Duolingo a way to distribute an inference-heavy consumer feature without accepting the margin profile it expected when Video Call launched. The next decision belongs to von Ahn: preserve Max by finding another premium capability, limit Video Call inside Super, or collapse the top tier as its original cost rationale disappears.