Free Steam voice game tops $1,000 daily AI bill and takes bank loan

Easy Fox, the developer behind the free Steam demo Teach My Little Sister How To Drive, says player interest has grown more than 20 times beyond its expectations. The game uses Google Gemini to recognize voice commands and build responses, with ChatGPT as a fallback system. Because the demo is free, the studio is paying the ongoing AI usage costs while players test it.

Costs climb after player count jumps

Easy Fox wrote in a Steam update that it is spending over $1,000 per day on AI services for the demo. The team said the increase is not because each player is particularly expensive, but because so many people are playing, and it has taken out a bank loan to help keep the demo running.

SteamDB listed an average player count under 10 until late August, when the title began gaining attention. By late September it was around 50 players, and it reached a peak of 431 on Oct. 1.

Fallback models create quality problems

The higher usage has pushed the game into limits on Google Gemini, forcing a switch to the less capable ChatGPT Realtime Mini 2.1 model. Easy Fox said the cheaper fallback models have injected slop into gameplay, with player complaints including delayed responses and wrong inputs. The studio is also testing alternative AI models for smoother gameplay in Russia, where it said access to Western models is limited by sanctions, and finding replacements could be difficult after Ukraine struck Russia's largest data center.

Easy Fox looks to local hardware for the release

Easy Fox plans to introduce local AI models for players with capable hardware so they do not have to rely on external services. For the paid release, the studio said it plans to account for expected AI usage in the game's price and that players will not be charged separately for individual AI token usage after purchase.

The update reflects a broader cost problem in AI services. Some companies have abandoned tokenmaxxing as AI agents use far more tokens than chatbots. Although AI prices are falling rapidly compared with other technologies, heavy agent usage has made the technology more expensive than anticipated, at a time when AI subscriptions have hit a pricing wall.

Sources: Tomshardware