cd /news/artificial-intelligence/google-says-gemini-4-argon-agents-be… · home › topics › artificial-intelligence › article
[ARTICLE · art-153176] src=mashable.com ↗ pub= topic=artificial-intelligence verified=true sentiment=· neutral

Google says Gemini 4 Argon agents beat Minecraft without human help

Google DeepMind researchers reported in an Oct. 15 paper that a team of four Gemini 4 Argon agents defeated Minecraft's Ender Dragon final boss in a freshly generated survival world on Easy difficulty, taking 27 hours and 14 minutes and dying 605 times, mostly to monsters. The researchers said it was the first time AI agents beat Minecraft start to finish using only standard tools with no game-specific human help, though they acknowledged the agents played through code with information a human player would not have, such as knowing blocks hidden behind other blocks. The run cost about $520 in model usage at Argon's introductory API price, and the researchers noted the human record for beating Minecraft in a random world is under seven minutes.

by read3 min views1 publishedOct 19, 2026
Google says Gemini 4 Argon agents beat Minecraft without human help
Image: Mashable (auto-discovered)

Chance Townsend Mother Jones.

Read Full Bio Google claims its newest AI model, Gemini 4 Argon, just beat Minecraft — no human help required.

In a research paper posted Oct. 15, Google DeepMind researchers said a team of four Gemini 4 Argon agents defeated the Ender Dragon, the game’s final boss. The agents started in a freshly generated survival world on Easy difficulty and were given a single goal: kill the dragon. It took them 27 hours and 14 minutes. Along the way, the AI players died 605 times, mostly to monsters.

The researchers say this is the first time AI agents have beaten Minecraft from start to finish using only standard tools, with no game-specific help built by humans. Researchers watched the run on a live dashboard and could have stepped in at any point, but they say they never needed to.

You May Also Like

However, Google is upfront about the limits. The game ran on Easy, which meant that monster attacks dealt less damage to the AI players. Because the AI agents play the game through code rather than on a screen, they have access to information that a human player would not. For example, the AI agents knew what blocks were hidden behind other blocks, making the "mining" part of the game rather insignificant. Plus, given Gemini 4 Argon's training base, the AI agents knew the optimal strategies for playing the game.

[Terms of Use](https://www.ziffdavis.com/terms-of-use)and

[Privacy Policy](https://www.ziffdavis.com/ztg-privacy-policy).

It's impressive, but nowhere near what humans can do. The researchers note that the record for beating Minecraft in a random world is under seven minutes.

One may ask, why make an AI play Minecraft? The researchers argue that the skills involved, like working as a team, bouncing back from mistakes, and sticking with a goal for more than a day, are the same ones AI agents need for real-world work. According to Google, the run costs about $520 in model usage at Argon’s introductory API price.

Making a frontier AI model play a video game is actually a common benchmark used by many companies when testing. The Pokémon games specifically have captured the zeitgeist in the AI industry for a while now, but Minecraft is now catching on too. As Google's research points out, two community projects have recently reported beating the dragon with AI, including one using Anthropic’s Claude Opus 5.

Want more tech and digital culture news delivered to your inbox daily or sent straight to your device? Sign up for Mashable's Top Stories newsletter or get Mashable push alerts today.

Chance Townsend is the General Assignments Editor at Mashable, covering tech, video games, dating apps, digital culture, and whatever else comes his way. He has a Master's in Journalism from the University of North Texas and is a proud orange cat father. His writing has also appeared in PC Mag and Mother Jones.

In his free time, he cooks, loves to sleep, and greatly enjoys Detroit sports. If you have any tips or want to talk shop about the Lions, you can reach out to him on Bluesky @offbrandchance.bsky.social, by email at [email protected], or on Signal at 224-417-6628.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @google 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/google-says-gemini-4…] indexed:0 read:3min 2026-10-19 · —