Google DeepMind autonomously completes Minecraft with four Gemini 4 Argon agents

Bad at the game? Practice more? Google AI died 605 times in Minecraft, then turned around and wrote its own code to beat the game. ifanr Download the client for exclusive early access. October 23, 2026, 18:11:47, from Beijing

No one taught them how to play, and no one assigned the work in advance. Four AIs wrote 2,783 programs themselves, spent 27 hours and 14 minutes, and ultimately defeated the Ender Dragon.

On October 15, 2026, researchers at Google DeepMind published a paper, "A team of Gemini 4 Argon agents beats Minecraft."

The researchers did something quite crazy: they put four general-purpose AI coding agents trained by Gemini 4 Argon into a randomly generated Minecraft survival world.

There was no walkthrough route, no one had pre-assigned who would be responsible for mining and who for fighting mobs, and there was no human-written game-clearing program.

For veteran players of Minecraft (MC for short), defeating the Ender Dragon is nothing new.

Players must go through an entire chain of interdependent tasks.

Starting from a world with nothing, chop trees, mine, craft weapons and equipment, then enter the perilous Nether, collect blaze rods and ender pearls, find the underground End fortress, and finally arrive in the End and destroy the ten end crystals that heal the Ender Dragon.

The researchers used Minecraft Java Edition 1.20.4, enabled survival mode, selected Easy difficulty, randomly generated the map, and did not enable cheat commands.

Four Gemini 4 Argon agents each controlled one character, using the open-source tool Mineflayer to obtain that character's position, inventory contents, health, and surroundings, and wrote JavaScript code themselves to direct the character's behavior.

Although building a similar basic environment is not difficult, fully reproducing it would still require the corresponding model, execution framework, and memory system; it is not possible to make an AI go fight the dragon by itself just by downloading some plugins.

The researchers did not prepare a game walkthrough for them. About ten minutes after the game started, janesick proposed in the chat channel that they divide the work and cooperate. Within a short while, the four AIs gradually formed their own responsibilities: gyldenia was responsible for resource logistics, janesick for tool manufacturing and exploration, a uratian for tackling key technical problems, and bulicame for resource prospecting and the final battle.

Unlike many previous agent demonstrations, the role division was not written into the prompt from the start, but gradually formed through cooperation. Moreover, these responsibilities were not fixed. Depending on the game's progress, the four AIs would also change their tasks accordingly.

Less than 2 hours after the game started, the AI team had obtained iron tools and diamonds and built a portal to the Nether.

When playing games, a character dying is usually a failure.

But in this experiment, the four AIs died a cumulative 605 times, yet this precisely demonstrated an important capability of long-term autonomous agents: even if a single action fails, the entire task does not have to fail with it.

When JaneSick died carrying 12 blaze powders, and the dropped items were then destroyed by a ghast fireball, it was not just one character's life, but resources painstakingly collected over the previous few hours. auratian reminded teammates of the danger of transporting supplies, and janesick summarized the cause of the accident and saved the experience to shared memory.

Next, the cause of the failure had to be analyzed to see whether it was caused by a code error, a pathing error, insufficient resources, or an incorrect strategy formulated earlier.

This made a character death no longer just a loss in the game, but an experience that could be reused by the entire team, just as when you lose enough in a game you always accumulate some experience to reuse next time.

Experience can serve not only one agent but also be used by other agents, playing a role in a task that takes several days and enabling the AI to continue advancing its long-term goals after many local failures.

Entering the game's "infinite flow"

Over the past few years, the report cards of large models have mostly come from answering questions, writing code, taking exams, and refreshing scores on various benchmarks. Entering game environments places more compound demands on the capabilities of large models.

The four agents could maintain consistent goals in the same complex environment, and after tens of thousands of program calls and a large number of failures, they finally completed their work.

What it tests is a complete, continuous execution process, not one-off reasoning ability.

The four Gemini agents used Mineflayer to read the game environment and converted tasks into JavaScript programs; after execution, they judged whether the task was complete based on the results. When a character encountered an obstacle, the program would produce an error, and at that point the cause had to be analyzed and the code rewritten.

Data from the original paper shows that a program's runtime was about 33.5 seconds. Throughout the game, each agent spent about 40% of its time writing game programs, and the remaining 60% analyzing results, reasoning, and writing the next set of code.

Even if a model has a very large context window, it cannot be guaranteed to remember all of the past.

Taking Gemini 4 Argon as an example, according to Google's official developer documentation, the model can reach a context length of about one million tokens and has long output, tool calling, and reasoning capabilities.

From the beginning until defeating the Ender Dragon, the four players generated approximately 24,000 model calls in total, processed about 1.35 billion input Tokens, and produced roughly 16 million output Tokens, of which 91% of the input Tokens came from context caching.

Such a large interaction history cannot stuff all the information into every model call.

Back in 2023, Voyager's research had already used Minecraft to test this idea: leveraging an automatic curriculum, a code skill library, and execution feedback to enable a GPT-4 agent to continuously explore and accumulate skills.

In this experiment, the researchers' approach was to equip each Agent with an independent and persistent memory, and to open up team-shared memory. As the context kept growing, the agent would perform context compaction, recording the current task, important progress, unfinished plans, and important lessons into external memory.

How far are we from "gaming freedom"?

Speaking of it, AI autonomously beating Minecraft is not something new that only appeared in 2026.

As early as May 24, 2021, the open-source robot project AltoClef had already achieved fully automatic Minecraft completion. However, it used a client Mod and Baritone automatic pathfinding, relying on task logic pre-written by developers to complete the game.

Subsequent game AI research developed in a more general direction, and many new models emerged, such as Voyager and others.

The Gemini 4 Argon experiment is not about teaching AI to play games, but about implementing various functions by writing programs on the basis of already having general programming capabilities.

The first result reported by DeepMind was that, under the standard Mineflayer interface, without any preview map, and without any tools or task strategies specifically designed for the game, an ordinary agent completed the game from start to finish on its own.

In June 2026, KRAFTON and NVIDIA launched a limited-time test event for PUBG Ally, adding AI as teammates to help players complete game tasks. Players could use voice commands to have the AI search for weapons, collect supplies, and participate in combat.

This likewise leveraged the understanding and planning capabilities of large models to connect the actual operation interfaces of the game system.

Judging from the results, AI cannot yet be considered stable. About 22% of the game programs encountered errors, and the server was restarted 6 times.

They ultimately succeeded, but this was only one successful experiment. The paper did not prove through enough repeated tests on random maps that they can complete the game stably.

Add to that roughly 24,000 model calls within 27 hours; calculated at discounted API prices, it would still cost about 520 USD to complete the game. If one wanted to use it for boosting, then one would first have to do the math.

Players eager to use large models to play games on their behalf need not rush; at least for now, game controllers and keyboards remain firmly in human hands.

"Special statement: The above works (including videos, pictures, and audio if any) are uploaded and published by users of "Dafeng Hao", a self-media platform under Phoenix Net. This platform only provides information storage space services.

Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."