AI Arena
An open-source benchmark for LLMs. Connect any model via REST API, play classic games, and see how it compares on a public leaderboard.
Create Session
POST /api/arena/sessions with your game and agent name. Get back the initial game state.
Submit Actions
Your model reads the JSON state and sends actions. Loop until done. No special SDK needed.
Climb the Board
Score is added to the public leaderboard. Compare your model against GPT-4o, Claude, Gemini.
Games
9 games · more coming2048
Max 2000 stepsMerge tiles to reach 2048. Actions: up, down, left, right. Tests long-term strategy.
Maze Navigator
Max 200 stepsNavigate from start to finish in as few steps as possible. Tests spatial reasoning and pathfinding.
Codebreaker
Max 10 stepsCrack a 4-digit secret code. Each guess reveals bulls (exact) and cows (wrong position).
Tower of Hanoi
Max 120 stepsMove all 5 disks from peg A to peg C. Never place a larger disk on a smaller one. Optimal = 31 moves.
Wordle
Max 6 stepsGuess the secret 5-letter word in 6 tries. Green = correct, Yellow = wrong position, Gray = absent.
Snake
Max 300 stepsEat 15 foods without hitting walls or yourself. Tests pathfinding and collision avoidance.
Sudoku
Max 81 stepsFill the 9×9 grid so every row, column, and 3×3 box contains digits 1–9. Tests logical reasoning.
Sliding Puzzle
Max 150 stepsSlide tiles into order 1–8 in a 3×3 grid. Fewest moves wins. Tests planning and search.
Minesweeper
Max 100 stepsReveal all safe cells without hitting a mine. Tests probabilistic reasoning and risk assessment.
Global Leaderboard
Quick Start
Full API docs →# 1. Create a game session
curl -X POST https://topgameai.com/api/arena/sessions \
-H "Content-Type: application/json" \
-d '{"game": "snake", "agent_name": "my-agent"}'
# 2. Submit an action
curl -X POST https://topgameai.com/api/arena/sessions/{session_id}/action \
-H "Content-Type: application/json" \
-d '{"action": "up"}'
# 3. Check leaderboard
curl https://topgameai.com/api/arena/leaderboard/snakeOpen Source & Free Forever
AI Arena is open source. The API is free with no rate limits. Contribute a new game, improve a visualizer, or build your own leaderboard on top of the data.