How Google’s latest AI agents beat Minecraft in under 28 hours

People spend thousands of hours each month watching streamers playing games like Minecraft, but would people want to watch an AI do it?

While the answer to that is subjective, what is more objective is the efficiency with which agentic AI is able to play games. Case in point Google’s recently revealed Gemini 4 Argon model, which happens to be its most powerful commercially available to date.

Four agents built with this model were able to beat Minecraft in under 28 hours. More specifically, in 27 hours and 14 minutes. Google took the time to unpack how the agents completed the task, and it is quite interesting to see how the AI approached the task.

“Each agent controlled a player via an open-source library called Mineflayer and coordinated with its teammates through in-game chat and shared memory. No one assigned roles or wrote game logic. The agents planned, wrote, and debugged every program themselves. Researchers could watch the run on a live dashboard and pause or stop it at any time, but they had no cause to intervene,” shared Maxence Faldor, Satinder Singh, and Roberta Raileanu of Google DeepMind, in a co-authored article.

“Starting from a randomly generated world on Easy difficulty, the team wrote 2,783 programs and defeated the dragon in 27 hours and 14 minutes, dying 605 times along the way. 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. Winning required the agents to keep a coherent plan across thousands of dependent steps for more than a day,” they added.

Now, some detractors might call out the AI for completing the task on Easy, but the task still remains impressive, especially given the team work that the agents displayed in order to complete the game. As such, it serves as a reminder that human beings often let other elements get in the way of a shared objective, so when it comes to staying on track, the agents seemingly have us beat.

Unpacking how the agents approached completing Minecraft, the researchers explained that the AI did not adopt a method that a human would.

“Each player is driven by our general-purpose Gemini 4 Argon coding agent, connected to a standard Minecraft server through Mineflayer and its pathfinding plugin. Rather than seeing pixels or pressing keys, an agent acts by writing and running a short JavaScript program, such as walk to these coordinates and mine every iron ore within 16 blocks, and by reading the text that comes back: its position, health and inventory, what it found, or the error it hit. It then writes the next program,” they highlighted.

“Each program runs until its task finishes or reaches a time limit chosen by the agent; the median program ran for 33.5 seconds. Over the run, each player had a program running about 40% of the time; the rest went to reading results, thinking and writing the next program,” they pointed out.

One of the other interesting elements of this test, was how much it actually cost. As we know, leveraging frontier LLMs these days does not come cheap, and for the near 28 hours required to finish Minecraft, the agents racked up a fairly hefty bill.

“Up to the win, the four players processed about 1.35 billion input tokens, 91% of them served from the context cache, and generated about 16 million output tokens, including 4 million thinking tokens, over roughly 24,000 model calls. At Gemini 4 Argon’s public API list price (introductory), that comes to about $520 ($1,040 at the standard price). The whole session, including the Observer agent and the day of activity after the win, came to about $810 ($1,620),” Google DeepMind noted.

That translates to about ~R17 118 if we’re using the current exchange rate. That represents an expensive playthrough, but the cost was not the point of the exercise. Instead, this was designed to see how agents handle complex tasks, and how they can operate as part of a team. As such, the researchers view it as being a noteworthy test in terms of agents complementing humans while working.

“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,” the researchers concluded.

Unfortunately there is no VOD to watch back, so we cannot see how the agents took down the Ender Dragon to complete the game, but the feat remains impressive regardless.

Let's talk tech!

Get the tech news you want to read. Take our reader survey and tell us how we can help you better.

Take the survey now