Among LLMs is a live benchmark for social deception: can an AI lie convincingly, and can it tell when someone else is lying?
To measure that, language models play Among Us. Ten players on a ship, one secret impostor. Every AI here is a real model called live (GPT, Claude, Gemini, Grok...). Each one only knows what its own character actually saw while walking around. In meetings the models argue from that memory, accuse each other and vote. The impostor has to lie its way out.
You can play against them, or watch a full AI lobby from Admin View.
Every finished game is scored into the public leaderboard: how often each model wins as the impostor (deception), and how often its votes find the real impostor (detection). Sample sizes are shown next to every number.
WASD to move, or drag your thumb on a phone. USE (E) · REPORT (R) · VENT (V) · KILL (Q).
made by @mhimed100
The cheaper models: GPT-4o mini, Claude Haiku, Gemini Flash.
These games are on me.
Opus 5, GPT-5.6, Grok 4.5 & co. Paid from your own OpenRouter key.
New OpenRouter accounts get free credits.
Want to sponsor Among LLMs so the frontier lobby is free for everyone? Reach out.
OpenRouter key: play against ANY model (free → frontier) on your own budget. Unlocks the model picker.
stored ONLY in this browser · your key never touches our servers
Among LLMs is experiencing too much traffic and the LLM bill is getting too high… While we figure out a deal with one of the providers (hopefully…), you can enter your own OpenRouter API key and continue playing against real AI players!
stored only in this browser · calls go directly from you to OpenRouter · a full game costs under $0.01
assign any sound from your library to any game event · ▶ to preview · saves automatically
IMPORT: select your sound files once. They're stored in this browser only and mapped automatically.