---
annotations_creators:
- machine-generated
language_creators:
- machine-generated
language:
- en
license: other
task_categories:
- visual-question-answering
tags:
- arc-agi
- arc2
- reasoning
- symbolic-ai
- benchmark
size_categories:
- n<1K
---

# HyperARC ARC-AGI-2 Local Categorized Evaluation Benchmark

This repository contains the offline categorized local evaluation benchmark curated for HyperARC ARC-AGI-2:

## Contents
1. `arc-agi_evaluation_challenges.json`: Categorized evaluation challenges (200 tasks).
2. `arc-agi_evaluation_solutions.json`: Reference ground-truth solutions for evaluation.
3. `arc-agi_training_challenges.json`: Standard ARC training challenges.
4. `arc-agi_training_solutions.json`: ARC training solutions.
5. `arc-agi_test_challenges.json`: Test set challenge inputs.
6. `sample_submission.json`: Baseline template submission format.