Easy Fox, the developer behind the free Steam demo Teach My Little Sister How To Drive, has taken out a bank loan to keep the game running. The demo relies on Google Gemini to process voice commands and generate responses, with ChatGPT as a fallback. Player numbers grew more than 20 times beyond expectations, driving AI service costs past $1,000 per day.
The team said the high cost is not because each player is expensive to support, but because so many people are playing. For the paid release, they plan to fold expected AI usage into the game's price rather than charging separately for tokens. They are also exploring local AI models for players with capable hardware, which would reduce reliance on external services that can hit usage limits.
The demo has already run into Gemini limits, forcing it to fall back to ChatGPT Realtime Mini 2.1, which caused delayed responses and wrong inputs. The developers are also testing alternative models for Russia, where sanctions limit access to Western AI services. The article notes that rising AI costs have led some companies to abandon "tokenmaxxing," since AI agents consume far more tokens than chatbots, making expenses higher than anticipated even as prices per token fall.