Skip to main content

Traditionally, to measure AI, static benchmarks have been the yardstick. These work well for evaluating LLMs and AI reasoning systems. However, to evaluate frontier AI agent systems, we need new tools that measure:

  • Exploration
  • Percept → Plan → Action
  • Memory
  • Goal Acquisition
  • Alignment

By building agents that can play ARC-AGI-3, you’re directly contributing to the frontier of AI research.

Learn more about ARC-AGI-3.

Human playing LS20

Can you build an agent to beat this game?

​
Play your first ARC-AGI-3 environment

​
1. Install the ARC-AGI Toolkit

uv init
uv add arc-agi
# or
pip install arc-agi

​
2. Set your ARC_API_KEY

Optionally set your ARC_API_KEY. If no key is provided, an anonymous key will be used. However, registering for an API key will give you access to public games at release. Get an ARC_API_KEY 
export ARC_API_KEY="your-api-key-here"
# or
echo 'ARC_API_KEY=your-api-key-here' > .env

​
3. Play your first game

Create a file called my-play.py: 
import arc_agi
from arcengine import GameAction

arc = arc_agi.Arcade()
env = arc.make("ls20", render_mode="terminal")

# Take a few actions
for _ in range(10):
    env.step(GameAction.ACTION1)

print(arc.get_scorecard())
Run it: 
python play.py
You should see the game render in your terminal and a scorecard with your results. 🎉 Congratulations! You just played your first ARC-AGI-3 environment programatically. Do you feel the AGI yet?

​
Next Steps

After running your first environment:
  1. Make it fast - Use env = arc.make("ls20") without render_mode to hit +2K FPS
  2. Try a different game - Run env = arc.make("ft09", render_mode="terminal") to play a another game. See a list of games available at three.arcprize.org or via the ARC-AGI Toolkit
  3. Use an agent - Explore agent templates or create your own agent.
  4. Explore the ARC-AGI Toolkit - The ARC-AGI Toolkit allows quick and easy integration with ARC-AGI Environments.