Beating AI News Flash: In AI model evaluation company Vals AI's Minecraft long-horizon test, GPT-6 Astra advanced to a point no AI system had ever reached before. It built a semi-automatic blaze farm, obtained 6 blaze rods, then found a warped forest, killed more than 6 endermen, and collected 3 ender pearls.
But Astra then put all its important supplies into a single chest. A creeper suddenly appeared and blew up the chest and the bed together, causing almost all key supplies to be lost.
After that, Astra spent several hours almost only farming potatoes, and began repeatedly staying on guard against creepers. It would even remind itself upon seeing anything green, "This is sugar cane, not a creeper."

