Vals AI, a company specializing in AI model performance evaluation, conducted a test using OpenAI's GPT-6 Astra to autonomously play the game "Minecraft." The purpose of this test was to investigate the ability of AI agents to operate a computer over an extended period.
Product LaunchesOpenAIVals AIGPT-6 Astra
GPT-6 Astra plays Minecraft autonomously for 141 hours, but fails goal due to Creeper explosion
This article is a translation. Read the Japanese original
In the test, the game was played through standard computer operations without the use of specialized control tools. GPT-6 Astra made steady progress toward the endgame, including reaching a Nether Fortress, constructing a semi-automatic device (Blaze Farm) to collect Blaze Rods (materials obtained from Nether monsters), and gathering Ender Pearls. According to Vals AI, its progression speed exceeded that of previous AI systems.
However, while attempting to store collected items in a chest, a problem occurred when a Creeper explosion caused the loss of items and the respawn point. Following this incident, progress stalled, and the model was observed engaging in activities such as farming potatoes for several hours.
Ultimately, GPT-6 Astra failed to defeat the final boss, the Ender Dragon, and was unsuccessful in clearing the game. Nevertheless, Vals AI evaluated that the model demonstrated higher gameplay capabilities than previous AI systems, given its ability to autonomously perform advanced item collection.