Google DeepMind
Google DeepMind has achieved another major gaming benchmark, announcing that a team of four autonomous artificial intelligence (AI) agents powered by its new Gemini 4 Argon model successfully completed Minecraft from scratch in just 27 hours. Twenty-seven hours and 2,783 custom programs later, the team conquered the game by slaying its final boss, the Ender Dragon. “Four Gemini 4 Argon agents beat Minecraft from scratch in 27 hours. Operating in a sandbox using only code and text, they wrote 2,783 programs, divided up jobs over chat, shared notes, and debugged errors together – all with no human help,” Google DeepMind said. The decisive final blow was struck by a surviving agent clinging to just 2.5 out of 20 health points.
Because it requires maintaining a long-term, coherent operational plan across thousands of dependent, high-stakes tasks, completing Minecraft in a randomly generated survival world represents one of the steepest tests in computer science.
Demonstrating that autonomous agents can collaborate, rebound from repeated failures, and stick to a multi-day objective proves they can tackle the open-ended problem solving required for cutting-edge software engineering, scientific research, robotics, and complex physical systems, according to DeepMind. Unlike DeepMind’s historic gaming breakthroughs, such as AlphaGo and AlphaStar, which relied on systems custom-trained specifically for a single title, the latest feat was accomplished using general-purpose coding agents – the identical software assistants engineers deploy for daily programming tasks. Operating inside a sandboxed survival world with human oversight limited strictly to safety monitoring, the four AI agents wrote code in real time, coordinated division of labor via text chat and debugged thousands of errors without any pre-programmed game logic or human intervention. Players must: Harvest and process raw materials, Venture into dangerous alternate dimensions to secure rare resources, Use spatial mathematics to locate a hidden underground fortress, and Eliminate a flying boss monster that continuously regenerates health. Any misstep can lead to death, stripping the player of inventory and forcing hours of recovery. Thirty-nine minutes after destroying the final healing crystal, the agents brought down the dragon. You use AI every day. Now get your AI Quotient. Take the AIQ test.
such as running on peaceful mode, exploring pre-scouted maps, starting with active portals, or using hard-coded game scripts, DeepMind sought to test whether standard coding models could accomplish the entire journey unaided, armed only with their general training data and a bare code interface While earlier community tests achieved partial runs by relying on shortcuts.