II.
Benchmark overview
Reference · livebenchmark:arc-agi-3
ARC-AGI 3 overview
Interactive visual-reasoning benchmark from François Chollet / ARC Prize Foundation (2025). Agents play short novel grid-based games to test fluid intelligence and pattern induction over interactive environments rather than static IQ-style puzzles.
Attributes
displayName
ARC-AGI 3
homepageUrl
kind
interactive-reasoning
targetsKind
AgentVersion
description
Interactive visual-reasoning benchmark from François Chollet /
ARC Prize Foundation (2025). Agents play short novel grid-based
games to test fluid intelligence and pattern induction over
interactive environments rather than static IQ-style puzzles.
Outgoing edges
applies_to2
- domain:computer-science·DomainComputer Science
- domain:ml-ai·DomainML/AI
covers2
- skill-area:visual-pattern-induction·SkillAreaVisual Pattern Induction
- skill-area:agentic-loops·SkillAreaAgentic Loops
Incoming edges
None.