JEV-Arcade
Fifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.
Overview
JEV-Arcade is a web arcade of fifteen game cabinets where language models play each other live, playable at jev-arcade.com or self-hosted with your own keys. Each cabinet shows one thing a judgment model does well. Up to four players per game can be Jev and any text model reachable through Anthropic, OpenAI, OpenRouter or local Ollama, and every move is shown with its reason or probabilities, tokens and cost, with cheats caught and logged.
The flagship is WikiRace: models start on a Wikipedia article and may only follow links on the current page to reach a target. Text models answer with a reason and a link, which lets them cheat by naming links not on the page or jumping to the target. Jev writes no text; each turn it answers one typed Choice over the links actually on the page, so it cannot cheat. The arcade floor has attract-mode loops, an LED results ticker and sounds, and remembers played cabinets in the browser. It has CI.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which of these links on the current page leads toward the target article? | — |
How it works
- Pick a cabinet and up to four players
- Server runs the game, asking each model for its move
- Jev answers a Choice over legal moves; LLMs answer in text
- Moves, costs and cheat attempts are logged live
Strengths
- Side-by-side comparison of Jev with text models
- Jev cannot make illegal moves by construction
- Shows tokens and cost per move
Caveats
- Self-hosting needs keys for each provider used
- Games demonstrate behaviour, not rigorous benchmarks
Getting started
Play at https://jev-arcade.com, or follow the Quick start to run with your own keys.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- web app with game server
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Fifteen live AI games: TypeSafe Jev vs.
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide in WikiRace?
Which link on the current page to follow next, as a Choice over real links.
Why can't Jev cheat?
It only picks from the links offered, while text models can name links that are not there.
Which other models can play?
Any model reachable via Anthropic, OpenAI, OpenRouter or Ollama.
Related projects
One-page web app where Jev plays chess against OpenRouter LLMs, Stockfish or you, showing the moves it weighed and their probabilities.
AppbetaJavaScriptPac-Man where Jev or other System One models choose each move at junctions, with live odds, fallbacks and a leaderboard.
ProjectbetaTypeScriptBrowser chess lab where Jev, LLMs and people play, each Jev turn being one Choice over the legal moves, with Stockfish analysis.
ProjectbetaPythonNiceGUI desktop chess app for evaluating Jev's move choices against Stockfish, random players or itself, with PGN export.
ProjectprototypePythonBrowser 2D driving sim where Jev answers lane, speed, hazard and pedestrian-yield questions every 200 ms, with an LLM comparison mode.
ProjectprototypeJavaScriptReal-time versus Tetris where Jev and an LLM or open Laya model pick placements on the same seeded pieces and clock.
ProjectprototypeTypeScript