Google DeepMind Team: Four Gemini 4 Argon Agents Beat Minecraft Fully Autonomously for the First Time
Four Gemini 4 Argon agents beat Minecraft autonomously without predefined roles, taking 27 hours to demonstrate long-horizon planning and collaboration.
ImportanceMaterialEvidenceE2 unreplicatedWrite-upQuick
Researchers from the Google DeepMind team dropped four AI agents running on Gemini 4 Argon into a randomly generated Minecraft world and successfully defeated the Ender Dragon without any human help. This marks the first time general-purpose large language models have fully autonomously completed the game without assigned roles, strategies, or game-specific tools.
The agents split up tasks on their own, warned each other about threats, recovered items from dead teammates, and logged mistakes in a shared memory. Over the course of 27 hours and 14 minutes, they wrote a total of 2,783 JavaScript programs, exchanged 229 chat messages, and died 605 times. The entire session cost $1,620 at the standard Gemini 4 Argon price that takes effect in 2027.
This result highlights progress in agentic AI within gaming environments. In recent years, beating complex video games like Minecraft typically required purpose-built AI systems trained from scratch, whereas current frontier general models can achieve similar goals through autonomous collaboration and code generation.