Pranav Innovates

Dev Tools Ranked · 5 Nov 2026

Someone Hid a Nintendo Inside an AI Model

Watch on YouTubeWatch on Instagram

About this video

nesformer puts the NES's processor, its memory and a game cartridge inside the weights of a stock Qwen3.5 AI model. Nobody trained it: a compiler calculated every weight. When you play, the model predicts the console's execution log one token at a time, and a small host program draws the picture. It's free and open source (MIT) and runs on a normal laptop CPU, about 200 times slower than a real Nintendo.

Filmed on my MacBook on 3 Nov 2026: Full Game by Doug Fraker (an MIT homebrew game), 460 frames (7.7 s of game time), 3,322,934 tokens, 26 minutes 22 seconds of real time.

Sources:

Not affiliated with Nintendo, Alibaba's Qwen team or the authors. Game: Full Game by Doug Fraker (MIT). Type: Geist and Geist Mono (Vercel, SIL OFL), Instrument Serif and Caveat (SIL OFL). Animated with Remotion.

Music: Hypnothis by Kevin MacLeod, CC BY 3.0, via Wikimedia Commons (https://commons.wikimedia.org/wiki/File:Hypnothis_(ISRC_USUAN1100634).mp3)

What the video says

Someone just hid a Nintendo inside an AI model. And it actually plays games. Here, let me show you. It's called nesformer. It's free, and it's open source. It starts as a normal Qwen AI model. But nobody trained it. A compiler calculated every single weight. The console's processor, its memory, even the game cartridge: all inside the model. Then you load a game, and the model predicts the console's next step, one token at a time. Every jump. Every coin. Every frame. Normally, an AI model guesses the next word. Here, it is the computer. The catch? It's slow. Around two hundred times slower than a real Nintendo. Eight seconds of play took my laptop twenty-six minutes. Comment NES, and I'll send you the setup guide with every command.

Topics

  • nesformer
  • Nintendo in an AI model
  • NES
  • Qwen
  • Qwen3.5
  • AI model weights
  • compiled transformer
  • Qwenputer
  • 6502
  • homebrew NES
  • open source AI
  • LLM