ARC-AGI-1
The short answer · 11 words
In short
Benchmark with visual grid tasks to detect and apply new rules.
Flow-alerta.cab (2026): ARC-AGI-1. https://flow-alerta.cab/en/glossar/arc-agi-1. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Helps to understand specialized AI benchmarks and their limitations.
Example scenario: A team investigates whether a model derives an unknown grid rule from the given examples.
Use as a research classification; additionally test own tasks for operational model selection.
At a glance
- Core
- ARC-AGI-1 includes tasks with input and output grids. From a few examples, a system has to infer the transformation and apply it to a new input. The specific task format examines generalization under limited examples; a result alone does not prove general intelligence.
- Use
- Helps to understand specialized AI benchmarks and their limitations.
- Example
- Example scenario: A team investigates whether a model derives an unknown grid rule from the given examples.
- Limits
- See high scores as AGI proof; mix training and testing tasks; confuse ARC with the AI2 Reasoning Challenge.
In context
ARC-AGI-2
Successor of ARC-AGI-1 with more demanding tasks for rule-based generalization.
Benchmark
Standardised task, test collection or measurement method used to evaluate systems under the most comparable conditions possible.
Generalization
Generalization refers to the performance of a model on new data not used for training.
Questions about this term
Does ARC-AGI-1 measure every ability?
No.
Is ARC always the same benchmark?
No, the abbreviation is also used for other tests.
All details and background
ARC-AGI-1 includes tasks with input and output grids. From a few examples, a system has to infer the transformation and apply it to a new input. The specific task format examines generalization under limited examples; a result alone does not prove general intelligence.
ELI5 – Explained simply
You see few solved puzzles and find the rule for another puzzle.
Why it matters in practice
Helps to understand specialized AI benchmarks and their limitations.
Example
Example scenario: A team investigates whether a model derives an unknown grid rule from the given examples.
Pitfalls and limits
- See high scores as AGI proof; mix training and testing tasks; confuse ARC with the AI2 Reasoning Challenge.
Origin & history
François Chollet presented the Abstraction and Reasoning Corpus 2019 in “On the Measure of Intelligence”. The original benchmark is listed as ARC-AGI-1.
Use in mid-sized companies
- Use as a research classification; additionally test own tasks for operational model selection.
Related terms
Matching Flow-alerta service
AI adoption & supportAI introduction and accompaniment.Matching tool / use case
View use casesUse Case: Specialised training on generalisation and benchmark boundaries.Sources
- arxiv.orgChollet, On the Measure of Intelligence (2019)
- arcprize.orgARC Prize, ARC-AGI-1 und 2
Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.
Editorially reviewed on: