Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

Four Gemini 4 Argon agents beat Minecraft's Ender Dragon with zero human help

Four Google DeepMind Gemini 4 Argon agents defeated Minecraft's Ender Dragon in 27 hours without human intervention, self-organizing roles and…

Four Gemini 4 Argon agents defeated Minecraft's final boss, the Ender Dragon, without any human intervention, completing the run in 27 hours and 14 minutes, Google DeepMind researchers disclosed in a new paper1,2. The run cost $1,620 at Argon's planned 2027 pricing.

The Google DeepMind team dropped the agents into a randomly generated survival world without assigning roles, strategies, or game-specific tools. Each agent controlled a player character through the open-source Mineflayer library and coordinated with teammates via in-game chat and shared memory.

During gameplay the agents elected on their own to split tasks. They warned each other about threats, shared coordinates, offered tips, and recovered items from dead teammates. Each agent settled into a distinct role: "janesick" made tools, "gyldenia" gathered food and supplies, "auratian" lit a key portal, and "bulicame" built Enderman shelters and landed the killing blow on the Ender Dragon. The majority of the 27-hour runtime was spent on the Ender Dragon fight itself.

The experiment is notable because the agents are general agentic AI models rather than game-specific software purpose-built for Minecraft. Previous AI systems that mastered complex games typically required custom architectures tailored to the target environment.

ANALYSIS The $1,620 cost figure, pegged to a price point that does not yet exist commercially, frames the demonstration as much as an inference-economics proof of concept as a capability showcase. The agents' ability to self-organize roles, share situational knowledge, and recover from teammate deaths without any pre-programmed coordination protocol is a concrete step beyond single-agent benchmarks toward multi-agent collaboration on open-ended tasks.