I asked precisely that question on their discord. An account of token spends.
Noting that this project is examplary experiment that may demonstrate whether the state of genAI can produce complex software before an explosion of token consumption as the code base grows, or even collapse.
In my experiments, it collapsed, each time I tried. Every 6 months or so I take the challenge. Unless a human intervenes of course. Basically dictating what genAi should do, which becomes more painful than it's worth automating code generation.