EVL
ARC-AGI
Abstraction & reasoning
What this measures
Abstract visual puzzles meant to measure *fluid intelligence* — spotting a transformation rule from a few examples and applying it. Famously hard for LLMs; a proxy for general reasoning.
Sample
Given 3 before/after colored-grid examples, infer the rule and produce the output grid for a new input.
Leaderboard · 0 models
% of puzzles solved exactly — higher is betterNo scores recorded yet.
What improves this score
Datasets & environments shown to raise it.
No data yet.
Research using this benchmark
all →SePO: Self-Evolving Prompt Agent for System Prompt OptimizationExplore Before You Solve: The Speed--Depth Trade-off in Epistemic Agents for ARC-AGI-3Interaction Locality in Hierarchical Recursive Reasoningoptimize_anything: A Universal API for Optimizing any Text ParameterGIM: Evaluating models via tasks that integrate multiple cognitive domains
Matched by name in title/abstract.