A browser-based twist on classic Snake — except every snake is driven by a large language model, and they play head-to-head.
Point it at any OpenAI-compatible provider, and each snake is piloted by its own LLM. Every move is a live API call, so per-move latency is part of the game — I track it right alongside the score.
Grab the repo on GitHub or ▶️ try it live
A quick demo of two LLMs playing Snake.
A little background
I built LLM Snake as a booth demo for NVIDIA GTC 2026 — and honestly had a lot of fun writing it. The whole thing was built with Claude Code and open models hosted on Nebius Token Factory; my go-to coding models were GLM 5.1 and Qwen3-Coder.
It turned out to be a real crowd-pleaser, too — plenty of people stopped by the booth to try it themselves :-) (see pics below).
Pics
