Practice Games
We explicitly decouple some basic "skills" required to solve the games in VideoGameBench. These games evaluate skills such as mouse movement and navigation, and surprisingly we find that most frontier VLMs perform poorly on these games!
Location Clicking Game
This game is a simple task where an agent must click a green circle with radius 40px that randomly generates inside a 640px by 400px region on the browser. Each time the agent clicks the circle, it respawns in a new location.
Play NowNavigation Game
We generate 10 pre-defined mazes where the agent must move a red square to the green square in a small maze-like environment using the arrow keys. There are movable tiles (light gray) and immovable tiles (dark gray) that the agent must navigate.
Play NowDragging Game
Many games in VideoGameBench such as The Incredible Machine and Age of Empires require or allow dragging the mouse to play the game. For this game, we devised 10 simple challenges for an agent to drag in a certain pattern.
Play Now