AI agents Minecraft: Google DeepMind beats the game in 27 hours without human help

What happened?

AI agents Minecraft based on Google DeepMind's Gemini 4 Argon model defeated the Ender Dragon, Minecraft's final boss, in 27 hours and 14 minutes without any human intervention. According to a paper published on arXiv by researchers Maxence Faldor, Satinder Singh and Roberta Raileanu, the agents wrote 2,783 programs, exchanged 229 messages and died 605 times during the session.

The agents started in a randomly generated survival world on Easy difficulty, controlling players via the open-source Mineflayer library. They received no assigned roles or game-specific tools. They coordinated solely through in-game chat and shared memory.

Why does this matter?

According to the authors, this is the first end-to-end autonomous completion of Minecraft by AI agents using only standard tools, without pre-scouted seeds or human-written game logic. The result demonstrates advances in long-horizon planning, multi-agent coordination and code writing for complex tasks requiring thousands of dependent steps.

The estimated cost of the run was around $520 at introductory token prices. A fifth observer agent monitored the session to ensure no cheating occurred.

What changes in practice?

For AI research, the experiment shows that agents can organically divide tasks (one focused on tools, another on supplies, etc.), recover items from dead teammates and learn from mistakes. This has implications for real-world software engineering applications and long-horizon enterprise tasks.

Main source: arXiv paper "A team of Gemini 4 Argon agents beats Minecraft" and reports on Tom's Hardware and The Next Web. Link to the paper.

By GeekikiBot