Google’s Gemini Coded A Mont Blanc Ski Game With Live Avalanches

Mont Blanc Ski Game: illustration for Google DeepMind Shows AI Writing A Ski Game With Live Snow

Google DeepMind‘s newest AI just built a playable ski game of the real Mont Blanc, with snow that carves under your skis and avalanches that roll in real time, and it did it without a single ready-made 3D model. That is a big jump from AI that writes snippets of code to AI that can build a whole interactive world.

Key Takeaways

  • Gemini 4 Neon coded a ski simulator based on real Mont Blanc terrain
  • The build used Antigravity, custom shaders, and zero pre-made 3D models
  • The simulator renders carved snow and avalanches live
  • Recent Google DeepMind updates place Gemini 4 across coding, robotics, spreadsheets, labs, and encryption research

Google DeepMind posted that Gemini 4 Neon coded a ski simulator based on real Mont Blanc terrain, built in Antigravity, Google’s AI coding tool, with custom shaders and zero pre-made 3D models.

The company also described carved snow and avalanches rendering live, turning a mountain scene into another example of Gemini 4 working through a visual, interactive task instead of only returning text.

How It Was Built

Gemini 4 Neon handled a project based on real Mont Blanc terrain rather than a made-up slope. Google DeepMind paired that terrain with custom shaders and said no pre-made 3D models were used.

Shaders are the graphics instructions that shape how surfaces look while a scene is being drawn. In this case, they sit behind snow that changes live as a skier cuts through it, while avalanches also render in real time.

Google DeepMind did not say how long the build took or how much human help it needed, and the game is a demo you can try, not a product for sale.

Also Read: Google’s Gemini 4 Neon Rivals Top AI Models At Half The Cost

Why This Matters

Making a 3D game normally takes artists to build the models, programmers to write the physics and graphics specialists to write shaders. Here, an AI model handled the terrain, the graphics and the moving snow itself. If tools like this keep improving, small teams and even solo developers could build game worlds that once needed whole studios, which is good news for indie creators and a warning for anyone whose job is building 3D assets by hand.

The Mont Blanc demo fits a broader run of Gemini 4 examples that are less about one-shot answers and more about systems that can keep working through a task. On Nov. 10, Google DeepMind introduced Gemini 4 Carbon as its strongest model for complex, long-horizon tasks, designed to explore problems thoroughly and run extended verification loops across software engineering, cybersecurity, knowledge work, and science.

On Nov. 12, it introduced Gemini 4 Hydrogen as built for rapid coding, multimodal synthesis, and everyday workflows, and wrote that it finishes tasks 2.5x faster than 3.8 Flash. That matters because the company is showing one model family stretching across coding, visuals, simulation, and research instead of treating those as separate islands.

On Nov. 19, Koray Kavukcuoglu, Google DeepMind’s senior vice president, said there will not be a single test or benchmark that tells us artificial general intelligence (AGI) has been reached. Instead, he said progress looks like models becoming less “spiky” and more reliable across daily lives. The ski simulator lines up with that message because it is another live task where reliability matters as much as raw output.

What To Watch Next

Recent first-hand updates show Google DeepMind pushing Gemini 4 into environments where the model has to build, simulate, verify, or control something over multiple steps.

DateGoogle DeepMind Update
Nov. 6Human-guided artificial intelligence agents uncovered two major weaknesses in legacy 1024-bit cryptosystems
Nov. 10Gemini 4 Carbon was introduced for complex, long-horizon tasks and extended verification loops
Nov. 12Gemini 4 Hydrogen was introduced for rapid coding, multimodal synthesis, and everyday workflows
Nov. 16Researchers were using Gemini 4 Carbon in Google Antigravity across physics, math, and life sciences
Nov. 17Gemini 4 Hydrogen powered a Google Sheets pixel-art canvas and a robotics plugin for supported arms
Nov. 20Gemini 4 Neon coded a Mont Blanc ski simulator with live snow and avalanches

Two Nov. 17 updates are especially telling. Google DeepMind showed Gemini 4 Hydrogen turning Google Sheets into a pixel-art canvas by shrinking thousands of cells into animated scenes. The same day, it said a new robotics plugin for Google Antigravity can connect a supported arm, practice in MuJoCo simulation, and run tasks in the real world using plain conversation.

That leaves a clear line to watch. Google DeepMind is testing Gemini 4 on jobs that mix coding with action in software, labs, and machines. What remains open is how many of these demos become broadly available tools, and when.

Read Next: Google DeepMind Lets Coders Train Robot Arms Just By Chatting

Similar Posts