Four AI agents running on Google’s Gemini 4 Argon have defeated the Ender Dragon, the final boss of Minecraft, without any human help. The run took 27 hours and 14 minutes from the agents’ first program, according to a paper Google DeepMind researchers posted on arXiv.
Did the AI agents get any human help?
The agents started in a randomly generated survival world on Easy difficulty, with cheat commands turned off. Nobody gave them roles, a strategy or game-specific tools. Their only goal was to beat the dragon.
They never saw the screen or pressed a key. Each agent controlled a player through Mineflayer, an open-source library for Minecraft bots. To act, an agent wrote a short JavaScript program, such as an order to walk somewhere and mine nearby iron. It then read back its position, health and inventory, or the error it hit.
Over the run, the four wrote 2,783 programs, exchanged 229 chat messages and died 605 times. The dragon accounted for 69 of those deaths.
How did the agents beat the Ender Dragon?
The agents sorted out the work themselves. One, “janesick,” made tools, found the Nether fortress and located the stronghold that holds the portal to the End. It threw two Eyes of Ender from different spots, tracked their flights in code and calculated where the paths crossed. The estimate landed 43 meters from the portal room, 1.3 kilometers away. “gyldenia” gathered food and supplies. “auratian” lit the Nether portal, and “bulicame” helped build the team’s Enderman shelters and later landed the final blow. The team kept shared notes on coordinates and lessons, and it recovered each other’s items after deaths.
The fight in the End lasted nearly 14 hours. About 21 hours in, auratian worked out that ten crystals on top of pillars were repeatedly healing the dragon. It also found the team could reach them by stacking dirt. The four destroyed all ten over the next four hours and 40 minutes. The dragon died 39 minutes later, with bulicame at 2.5 of its 20 health points.
How much did the run cost?
The win was not cheap. Up to the victory, the agents processed about 1.35 billion input tokens. At Gemini 4 Argon’s introductory list price, that comes to about $520, or $1,040 at the standard price, the researchers estimate. The full session, including a fifth agent that watched for cheating and found none, came to about $810, or $1,620.
The paper also lays out the limits. Mineflayer tells the agents exactly which blocks and creatures are around them, even when they sit behind solid rock, and a plugin plans their routes. The run used Easy difficulty, and 22 percent of the agents’ programs ended in an error. The result also comes from a single run. Two community projects have reported earlier dragon kills, but those relied on game-specific tools or a fixed, easier setup, the paper notes.
The agents operated in a sandboxed world, and their programs could control only their own player. The researchers could pause or stop the run from a live dashboard but did not need to. Google DeepMind plans to use the same setup for coding, science and robotics work, Faldor wrote on X. Last week, cofounder Sergey Brin said Gemini now helps build the next Gemini.
Este artigo foi escrito com a assistência de IA.
News Factory APP - notícias agênticas para impulsionar seu SEO e AEO.