Google Gemini 4 AI Agents Beat Minecraft's Ender Dragon in 27 Hours With No Human Help

Google Gemini 4 AI Agents Achieve World-First Minecraft Victory

Google DeepMind researchers have demonstrated a significant milestone in AI development as four Google Gemini 4 Argon AI agents autonomously defeated Minecraft's final boss without human intervention. The general agentic AI models completed the game in 27.23 hours, marking what researchers believe is a world-first achievement in end-to-end autonomous gameplay1

1

2

2

.

The experiment placed four AI agents into a randomized Minecraft survival world with a single objective: defeat the Ender Dragon. No roles, strategies, or game-specific tools were assigned. Each agent controlled a player using the open-source Mineflayer library and coordinated with teammates through in-game chat and shared memory. Researchers could monitor the session via a live dashboard but had no cause to intervene throughout the entire playthrough2

2

.

How AI Agent Coordination Enabled Success

The agents quickly self-organized into specialized roles without explicit programming. Agent "janesick" focused on crafting items and making tools, "gyldenia" concentrated on gathering resources and food supplies, "auratian" handled critical tasks like lighting the portal, and "bulicame" built Enderman shelters for farming materials and ultimately landed the final blow against the Ender Dragon1

1

.

Throughout the 27.23-hour session, the four agents exchanged 229 messages, wrote and executed 2,783 programs, and died 605 times. They warned each other about threats, shared combat tips and lessons, provided coordinates for navigation, and even recovered items from dead teammates. A fifth Observer agent monitored the entire session as an independent witness to check for cheating and publish live summary reports1

1

2

2

.

Technical Achievement and Resource Requirements

Reaching The End required approximately 13.5 hours, with the remaining time spent battling the final boss. The agents had to complete complex prerequisite tasks including obtaining Ender Pearls and Ender Rods, crafting Eyes of Ender, tracking down a Stronghold, and completing puzzles—all requiring progression through basic Minecraft tool tiers2

2

.

The operation consumed 1.35 billion input tokens, costing $810 at Gemini's list price or $1,620 at standard pricing—effectively $15 per hour per agent. While the AI agents completed the game, they're far from speedrun territory: the current Minecraft speedrun record stands at six minutes and 36 seconds2

2

.

Why This Development Matters for AI

"To our knowledge, this is the first end-to-end autonomous completion of Minecraft by AI agents, achieved using only the standard Mineflayer library without pre-scouted seeds or other game-specific tools written by humans," the research team stated. "Winning required the agents to keep a coherent plan across thousands of dependent steps for more than a day"1

1

.

This achievement demonstrates how general agentic AI models can now tackle complex games that previously required custom AI software. Unlike Google DeepMind's famous AlphaGo AI, which was specifically designed to win at Go, these Gemini 4 agents used no game-specific programming to killed the Ender Dragon1

1

.

Researchers emphasized they wanted to study how teams of agents coordinate in transparent settings to understand such systems before deploying them for real-world tasks. "The skills required to solve Minecraft such as coordinating a team, recovering from failure, and pursuing a goal for more than a day, are also what agents need to work reliably alongside people," they explained2

2

. This suggests the technology could soon handle long-term goal-driven tasks in professional environments where multi-agent collaboration proves essential.