Duolingo CEO Luis von Ahn says the company’s AI Video Call feature “was costing about 30 cents per call, just generally on AI inference” at launch. Von Ahn says Duolingo has “managed to reduce that cost to well under one cent per call”, “mainly by moving to open source models or open weight models”.
Von Ahn made the remarks on Duolingo’s third-quarter 2026 earnings call on 3 November, as transcribed in posts on X; the clip recirculated on 9 November in a post by Tanay Jaipuria. By our arithmetic, $0.30 against under $0.01 is a cut of more than 30 times per call.
Business Pill · Open Weights
A short Business Engineer explainer on renting a model through an API versus running open weights yourself. It teaches the general idea only and does not discuss Duolingo.
The key insight: Duolingo CEO Luis von Ahn says on the Q3 2026 earnings call that the AI inference cost of a Video Call fell from about 30 cents at launch to well under one cent, mainly by moving to open source or open weight models. By our arithmetic, that is a cut of more than 30 times per call.
On the call, von Ahn tied the cost cut to moving Video Call into Super: “So we feel really good about that. And that allowed us to put video call in Super”. Video Call had been the feature of Duolingo Max, the company’s highest subscription tier.
Duolingo’s Q3 2026 shareholder letter, filed with the SEC on 3 November, says: “As we continue driving our per-unit AI costs down, we’re also making progress in expanding access to Video Call”. The letter says Duolingo began giving unlimited Video Call to new Super subscribers last quarter and is “on track to make it available to all Super subscribers by the end of this year”.
The letter adds that Duolingo has “decided to sunset Duolingo Max” and will move existing Max individual subscribers to a Super family plan at the same price.
AI inference cost per Duolingo Video Call: about $0.30 at launch and well under $0.01 now, as stated by CEO Luis von Ahn on the Q3 2026 earnings call on 3 November 2026. The more than 30 times cut is FourWeekMBA’s arithmetic.
What Was Posted on X
Duolingo CEO @LuisvonAhn on the costs of their AI video call feature:
“When we first launched it was costing about $0.30 per call, just generally on AI inference. We’ve managed to reduce that cost to well under $0.01 per call, and that’s mainly by moving to open-source models or open-weight models.”
Tanay Jaipuria posted on 9 November 2026 a quote from Duolingo CEO Luis von Ahn on the cost of the AI video call feature: about $0.30 per call at launch, now well under $0.01, mainly by moving to open-source or open-weight models. Von Ahn made the remarks on Duolingo’s Q3 2026 earnings call on 3 November; this piece checks them against the Q3 shareholder letter.
What the Letter Says on Margin
The letter reports a gross margin of 72.9% in the third quarter, against 72.5% a year earlier, and says the outperformance was “primarily due to lower feature-based AI costs and favorable hosting cost trends”. Revenue rose 14% to $309.0 million and paid subscribers rose 14% to 13.1 million.
Duolingo guides gross margin to 72.0% in the fourth quarter, down 90 basis points, and the letter says that is “as we expand the rollout of AI-based features on our platform”.
As we read the remarks, von Ahn links the cost cut to who gets the feature: Video Call moves from Max into Super, which Duolingo says will reach all Super subscribers by the end of the year.
As we read the letter, Duolingo sets the marginstory at the level of feature-based AI costs and hosting, not one feature, and guides gross margin lower in Q4 as AI features roll out wider.
As we read the Revolut and Duolingo numbers together, two companies have now put a figure on a switch to open models, one on coding agents and one on a consumer feature.
Luis von Ahn, Duolingo CEO, on the Q3 2026 earnings call, 3 November 2026, as transcribed on X
“We’ve managed to reduce that cost to well under one cent per call. And that’s mainly by moving to open source models or open weight models.”
Three Implications
FOR CONSUMER AI PRODUCTS Von Ahn says the cost cut allowed Duolingo to put video call in Super, a lower tier than Max.
FOR SUBSCRIPTION TIERS The Q3 letter says Duolingo will sunset Max and move existing Max individual subscribers to a Super family plan at the same price.
FOR MARGINS The letter reports gross margin of 72.9%, up from 72.5%, and guides 72.0% for Q4 as AI-based features roll out more widely.
The analysis says: “Every technology stack has layers. Some are scarce, some are abundant.”
As we read it, von Ahn’s remarks place inference for Video Call on the abundant side: Duolingo moved to open models. The analysis predates the call and does not discuss Duolingo.
What Is Not Established
The quotes come from transcriptions of the call posted on X; we did not have the official transcript. In the remarks we read, von Ahn did not name the models, the hosting set-up or the total Video Call spend. The more than 30 times figure is our arithmetic on von Ahn’s rounded numbers.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Von Ahn says Duolingo cut the AI cost of a Video Call from about 30 cents to well under one cent, mainly by moving to open source or open weight models, and that the cut let the feature move from Max into Super. The Q3 letter reports a 72.9% gross margin and credits the outperformance primarily to lower feature-based AI costs and favorable hosting cost trends.
A note on sourcing. Quotations of Luis von Ahn are from transcriptions of Duolingo’s Q3 2026 earnings call posted on X on 3 and 9 November 2026; margin, revenue and subscriber figures are from Duolingo’s Q3 2026 shareholder letter filed with the SEC, read on 9 November 2026. We did not contact Duolingo. The more than 30 times figure is our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
