{
  "id": 15737312,
  "title": "Google DeepMind claims four AI agents beat Minecraft in 27 hours by writing 2,783 programs of their own",
  "url": "https://urgent.news/2026/10/20/google-deepmind-claims-four-ai-agents-beat-minecraft-in-27-hours-by",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-20T10:25:00.000Z",
  "source": {
    "name": "Tom's Hardware",
    "slug": "tom-s-hardware",
    "url": "https://www.tomshardware.com/tech-industry/artificial-intelligence/google-deepmind-claims-four-gemini-4-agents-beat-minecraft-in-27-hours-by-writing-2-783-programs-of-their-own-the-ai-team-died-605-times-on-easy-difficulty-and-ran-up-an-estimated-usd520-in-tokens"
  },
  "original_language": "en",
  "account": "In a test involving four AI agents from Google DeepMind's Gemini 4 Argon model, the team successfully completed the game of Minecraft in 27 hours and 14 minutes, with the agents writing 2,783 programs on their own. The agents faced challenges, including 605 deaths and spending about half the run battling the Ender Dragon, the game's final boss. The agents were given the sole objective of defeating the Ender Dragon, and they achieved this by stacking dirt to reach the boss's crystals. The agents demonstrated teamwork and coordination through in-game chat and shared notes, with roles dividing among them without human intervention. The agents wrote short JavaScript programs, with 42 helper functions kept alive between programs. The experiment aimed to showcase the model's capability for complex, long-horizon workflows and agent coordination, highlighting how AI agents can handle intricate tasks such as decompiling games into source code.",
  "summary": "Google DeepMind researchers say four Gemini 4 Argon agents beat Minecraft in 27 hours, writing 2,783 programs with no human help.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}