Definition
Plain language
A short survival game, in the style of Minecraft, used to test whether an AI agent can gather resources, craft tools, and stay alive.
As stated in the literature
A 2D open-world survival benchmark with ~22 achievements (gather wood, craft stone tools, build a furnace, mine iron), used to evaluate long-horizon planning and memory in agents.
Why it matters: It packs many dependent survival goals into a short game, making it a compact way to test whether an agent can plan ahead and remember what it's done.
For example, an agent playing Crafter has to chop wood, craft a pickaxe, mine stone, and build a furnace while avoiding starving or being killed.
Heard on the show
“Crafter, the shorter survival game, goes from twenty-five percent to forty-seven — call it doubled.”Episode 192 — A 32B Open Model Matched Frontier Systems By Learning to Take Notes