Project detail · Games & Play
STICKBLADE ARENA — Physics-Based LLM Benchmark
Watch two randomly paired LLMs sword-fight in a physics arena and blind-vote which performed smarter.
stickblade-arena.vercel.app
Try live demo Live checked2 hours ago
Try this first: Start a match to see two LLMs sword-fight, then vote who's smarter.
- Open source
- Yes
Tech Profile
- Built by
- 2–5 people
- Time
- Ongoing
- Open source
- Yes
Source
I built a site where two LLMs sword-fight in real physics and you blind-vote who's smarter Two months ago I thought: what if the AI benchmark was just… watching them fight. So: https://stickblade-arena.vercel.app Two random LLMs get stickman bodies in a pymunk physics arena They each turn output JSON moves (swing, block, dash, shoot bow, throw flail…) Ragdolls, momentum, weapon collisions, the whole bit You watch the replay without knowing which model is which and vote who f
For the maker: hang the plaque, claim the project
This project is unclaimed — sign in, hang the plaque on your site, and it's yours.
For the maker: hang the plaque, claim the project
This project is unclaimed — sign in, hang the plaque on your site, and it's yours.
Keep exploring
Similar projects
Sign in to report a problem with this project.
Comments
Sign in to comment.