Back to Gap Map Model components / Benchmark / eval datasets

AI2 ARC

Allen Institute for AI
open / Overall score: 4.0(strong)

Benchmark of 7,787 grade-school science questions assembled to study advanced question answering, split into an Easy set of 5,197 and a Challenge set of 2,590. Published by the Allen Institute for AI in 2018, it remains a standard multiple-choice reasoning evaluation in model release reports.

Openness

5 high confidence
5.0
data
open(7,787 questions, downloadable on HF)
license
CC-BY-SA-4.0(open data license, redistributable)
datasheet
yes(HF dataset card + arXiv paper)
access
public(train/val/test splits all released, not held-out)

Fully open under CC-BY-SA-4.0 with dataset card; the canonical open-data benchmark profile.

  • https://huggingface.co/datasets/allenai/ai2_arc recorded 2026-09-29

    The dataset page lists the license tag "license:cc-by-sa-4.0", embeds repo state "gated":false, and its card describes "genuine grade-school level, multiple-choice science questions, assembled to encourage research in" question answering; the card's Licensing Information section itself reads "More Information Needed", so the license is stated only in the card metadata tag.

Adoption

4 high confidence
4.0

Adoption is measured through Hugging Face downloads of allenai/ai2_arc. ARC is a widely cited standard reasoning benchmark, appearing routinely in frontier model release reports, and this category counts dataset downloads on a far smaller scale than it does for software or models.

Capability

not assessed

Capability is not a meaningful axis for an eval dataset, so it is left unscored.

  • https://huggingface.co/datasets/allenai/ai2_arc recorded 2026-09-29

    Dataset card summary: "genuine grade-school level, multiple-choice science questions, assembled to encourage research in" question answering, split into a Challenge Set and an Easy Set; searched the body for accuracy, SOTA, non-commercial and held-out: none, and no performance or feature claim.

Verified 2026-09-27