Skip to main content
ARC-AGI-3 games are turn-based systems where agents interact with 2D grid environments through a standardized action interface. Each game maintains state through discrete action-response cycles.
  • Each response contains one or more 2D frame arrays plus game-state metadata.
  • Agents choose from the action IDs in the response’s available_actions field. ACTION6 also requires x and y coordinates.

Available Games

To see which games are available, either go to arcprize.org/tasks or make an API call to list games. Example games include
  • ls20 - Agent reasoning
  • ft09 - Elementary Logic
  • vc33 - Orchestration

Game ID

Game IDs are formatted as <game_name>-<version>. game_names are stable, but version may change as games update.

Grid Structure

  • Dimensions: Maximum 64x64 grid size
  • Cell Values: Integer values 0-15 representing different states/colors
  • Coordinate System: (0,0) at top-left, (x,y) format

Game Available Actions

Each game provides an explicit set of actions that an agent can take. Actions available vary per game. Typically, the available actions include:
  • Actions 1–4: ex: move up, down, left, or right
  • Action 5: a game-specific interaction
  • Action 6: A complex action (if supported by the game)
  • Action 7: undo
To learn more about each action and what it does, please visit the Actions.

Running a Full Playtest

To run a complete playtest, you’ll need to integrate your agent with scorecard management and the game loop. This is what happens “under the hood” when using an agent harness. Below is an example of the key steps. For a ready-to-use implementation, see the benchmarking harness.

Game State Enumeration

Full Playtest

This bare-bones example is also available as a notebook.
This workflow ensures your plays are tracked officially. For parallel playtests across games, use the benchmarking harness to handle the orchestration automatically.