What is AGITest?
AGITest is a benchmark designed to measure an AI's general intelligence by testing its ability to solve novel problems it hasn't seen before. Unlike traditional AI tests that focus on specific skills, AGITest evaluates how efficiently an AI can learn and adapt to new challenges - just like humans do!
Hmm, I see a pattern!
Key Features of AGITest
Pattern Recognition
Tests AI's ability to identify visual patterns from colored grids and generate correct solutions.
Efficiency Measurement
Evaluates not just problem-solving ability but also how efficiently the AI uses resources.
Human Baseline
Compares AI performance against human problem-solving abilities, with humans averaging 60% accuracy.
Humans vs. AI
Humans
Average accuracy on AGITest
Top AI Models
Current accuracy on AGITest
Even the most advanced AI models struggle with AGITest, highlighting the gap between artificial and human intelligence.
AGITest Evolution
ARC-AGI-1 Introduced
François Chollet introduces the first version of the test in his paper "On the Measure of Intelligence".
First AI Success
OpenAI's o3 model becomes the first AI to match human performance on ARC-AGI-1.
ARC-AGI-2 Released
A more challenging version is released, focusing on efficiency and preventing brute-force solutions.
Try a Sample Puzzle!
Can you figure out the pattern and solve this AGITest puzzle?
Input
Output
Hint: Think about reflection and symmetry!
Solution: The output grid is the input grid flipped horizontally (mirrored).
Fun Facts About AGITest
100% of AGITest puzzles can be solved by at least 2 humans in 2 attempts or less.
The Arc Prize 2025 contest offers over $725,000 in prizes for AI systems that can solve AGITest.
AGITest focuses on "core knowledge priors" - cognitive building blocks present from birth or acquired very early in human development.