Easy Fox is spending over $1,000 a day on AI to keep its free Steam demo running, and took a bank loan

The four-person studio behind Teach My Little Sister How To Drive says demo players grew more than twentyfold in a month, Gemini hit usage limits, and it may have to close the demo early.

AI-drafted from 4 sources. This story was written with AI from the outside reporting listed under Sources, and checked by automated tools before it went up. The originals have the full detail.

Easy Fox, the studio behind Teach My Little Sister How To Drive, says it is spending more than $1,000 a day on AI services to keep the game's free Steam demo online, and has taken out a bank loan to pay for it. The figures are in a development update posted with version 1.2.9 on 7 October.

The game is controlled only by voice. You sit in the passenger seat and talk your little sister through a driving lesson, and an AI model listens to you and answers in her voice. "Over the past month, the number of demo players has grown more than twentyfold," the studio wrote. "We are currently spending over $1,000 per day on AI services for the demo." It adds that this "isn't because each player is particularly expensive to support. The total cost has increased because so many people are playing, and we have taken out a bank loan to help keep the demo running."

The team is four people, and the post warns that "financial pressure may mean we have to close the demo earlier than expected before the full release. We have not made that decision at this time." The Steam page lists the full game for January 2027. Easy Fox says the paid version will build expected AI use into its price, and players "will not be charged separately" for their token use after buying it.

Rate limits and a fallback model

The studio's AI information page, last updated on 27 August, names Google's gemini-2.5-flash on Vertex AI as the primary model, with OpenAI's ChatGPT Realtime Mini 2.1 and ChatGPT Realtime Mini as fallbacks. In mainland China the game uses Alibaba Cloud and BytePlus instead. The page says voice goes straight from the player's machine to the provider, and the studio's own server only checks ownership and hands out a short-lived token.

The growth broke that order. Since the surge, Easy Fox says, Gemini "has been frequently returning usage-limit errors". It asked Google Cloud to raise its usage tier, "but were informed that a manual increase was not available", because the studio has to build up usage to qualify and the limits stop it doing so. Patch 1.2.6 on 29 September made the fallback the primary model. Players have since complained of long replies, stilted dialogue and the sister turning right when told to turn left, problems the studio puts down to leaning on the OpenAI model. Players in Russia, where the global version's AI services are blocked, get a separate, slower model.

The plans in the update are to get Gemini working reliably again, to add local models as an option for players whose graphics cards can run them, and to test other models for Russia.

Russian site PlayGround.ru reported on 10 October that the demo averaged fewer than ten players at a time until the end of August, about 50 by the end of September, and peaked at 431 on 1 October. Steam's own counter showed 146 people in the demo on Monday morning. The game was among the eight finalists for Sense of Wonder Night at Tokyo Game Show 2026, where it was also shown in the Indie 80 area.

Sources

Image: Easy Fox

More news from Gaming World