GPT-6 Astra Minecraft Run Advances Further Than Prior AI Tests Before Creeper Loss Stalls Progress

GPT-6 Astra advanced further in a 141-hour Minecraft evaluation than any AI system previously tested by Vals AI, building a semi-automatic Blaze farm and collecting late-game resources before a Creeper explosion erased key progress. The run became a vivid stress test for long-horizon computer-use agents: the model recovered poorly after losing its chest and bed, spending several hours farming potatoes and repeating cautionary reminders about storage and hostile mobs.