ITG NowHome › AI glossary › ARC-AGI
What is ARC-AGI?
ARC-AGI is a set of visual puzzles that tests whether an AI can learn a new pattern from just a few examples.
- Category Benchmarks
- Also called ARC-AGI-2, ARC Prize
- Updated 20 Oct 2026
ARC-AGI, explained simply
ARC-AGI shows small grids of colored squares. The AI must work out the rule from a few examples and apply it to a new grid.
People find most of these puzzles easy. AI models have found them much harder.
This matters because the test aims to measure learning new things, not memory. That is one idea of general intelligence.
Higher is better. The ARC Prize is a contest that rewards progress on it.
On ITG Now, the AI models page shows ARC-AGI-2 results, with a link to the source.
Related terms
BenchmarkAGIHumanity's Last ExamReasoning modelBenchmark contamination
Where you'll see it on ITG Now
- AI modelGPT-6 Astra
- AI modelClaude Fable 5.1
- AI modelClaude Haiku 5.5
- AI modelClaude Opus 5.5
- AI modelDeepSeek-V4.1-Flash
- AI modelGPT-6 Luna
Questions people ask about ARC-AGI
Is a higher ARC-AGI score better?
Yes. A higher score means the AI solved more of the puzzles.
What is the ARC Prize?
It is a contest that gives rewards for strong results on the ARC puzzles. Its site explains the rules.
Does solving ARC-AGI mean we have AGI?
No. The makers say it is one useful step. It doesn't prove general intelligence.