ARC Challenge (PapersWithCode): leaderboard
AI2 Reasoning Challenge (Clark et al., 2018), Challenge set: grade-school science multiple-choice questions that both a retrieval-based and a word co-occurrence baseline answered incorrectly. This board is the archived Papers With Code table of accuracies as each paper reported them, under that paper’s own setup (shot counts differ by row), not a common evaluation run; higher is better.
Metric: Accuracy. Source: huggingface.co. Saturation forecast: Estimated already saturated. 54 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 96.4 | #459 |
| 2 | GPT-3.5 | 85.2 | #825 |
| 3 | OPT-175B | 43.94 | #1507 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=arc-challenge-paperswithcode · How It Works · Data refreshed daily, snapshot 2026-11-05.