Free Steam demo runs up $1,000 daily AI bill and forces devs to take out bank loan

Easy Fox, the developers behind a free, AI-powered game demo on Steam, has been forced to take out a bank loan as more players tried out its title. The game, called Teach My Little Sister How To Drive, uses Google Gemini to listen to your voice commands and generate responses, with ChatGPT serving as a fallback system. In its latest update, Easy Fox said that the number of players trying out the game has grown over 20 times, which is more than what it expected. This led its costs to skyrocket to more than a thousand dollars daily, and cheaper fallback AI models have led to slop injected into gameplay, impacting the user experience.

SteamDB showed an average player count of under 10 people until late August, when it started gaining steam. By late September, this had ballooned to around 50 players, before finally hitting a peak of 431 on Oct. 1. “We are currently spending over $1,000 per day on AI services for the demo,” the team wrote on its Steam update. Easy Fox also added, “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.”

Since the game’s sole control method is your voice, AI is a critical part of its programming to recognize your commands and translate them into instructions for the virtual little sister. Since the game is still a free demo, Easy Fox needs to pay for the ongoing usage costs of the game. However, the developers said that the cost of using AI will be included in the price of the game once it gets final release. “For the paid release, we plan to account for expected AI usage in the game’s price,” the team said in its update. “Players will not be charged separately for their individual AI token usage after purchasing the game.”

The developers also said that they plan on introducing local AI models for gamers with capable hardware, so they don’t have to rely on external services, which could run into limits. This is one of the issues that the game has been grappling with, especially as the increased number of users has caused it to hit usage limits on Google Gemini. This has forced the game to fall back on the less capable ChatGPT Realtime Mini 2.1 model, which has led to some player issues, like delayed responses and wrong inputs. It also said that it’s testing alternative AI models for smoother gameplay in Russia, which does not have easy access to Western models because of sanctions. They might find it hard to find suitable replacement AI models, though, because Ukraine has struck Russia’s largest data center.

Increasing AI costs is one of the reasons why some companies have abandoned tokenmaxxing. Even though the price of AI has been falling quite rapidly compared to other technologies, the fact that AI agents use so many more tokens than AI chatbots has made the use of the technology more expensive than anticipated, especially as AI subscriptions have hit a pricing wall.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.