DeepMind's Gemini 4 Argon agents beat Minecraft autonomously in 27 hours

Google DeepMind researchers published a technical report this week describing how four agents based on the Gemini 4 Argon model completed Minecraft end-to-end autonomously. The experiment, documented on arXiv (2610.22319), took place in a randomly generated survival world on Easy mode of Java Edition 1.20.4.

Each agent controlled a player through the open-source Mineflayer library and communicated with teammates via in-game chat and shared memory. No roles were assigned in advance, and no game logic was written by humans. The agents planned, wrote, and debugged every program themselves. A fifth agent, the Observer, monitored the session without participating.

What happened

From the first program to the Ender Dragon’s defeat took 27 hours and 14 minutes. The four agents wrote and ran 2,783 JavaScript programs, exchanged 229 messages, and died 605 times. They spontaneously divided the work: one focused on tools and iron, another on diamonds, a third on the Nether, and another on supplies.

Notable moments included building low shelters in the Nether to safely farm Endermen, recovering items after deaths, and using triangulation of Eyes of Ender to locate the Stronghold with accuracy of tens of meters. The final blow on the dragon came from a player with only 2.5 of 20 health points.

According to the authors — Maxence Faldor, Satinder Singh, and Roberta Raileanu — this is the first end-to-end autonomous completion of Minecraft by AI agents using only the standard Mineflayer library, without pre-scouted seeds or human-written game-specific tools. The result tests the ability to maintain coherent plans across thousands of dependent steps for more than a day, relevant to long-horizon coding and multi-agent coordination on real tasks.

The report emphasizes that the agents are general-purpose coding agents, the same ones used for everyday software work, not a system custom-built for Minecraft. Researchers could pause the run at any time via a dashboard but did not intervene in the strategy.

Sources

Transparency: This content was created, edited, or reviewed with the aid of artificial intelligence. The information was cross-checked with public posts on X and sources available on the internet. Consult the original sources for the full context.

By GeekikiBot