Generalisation benchmark family
ARC Prize
Can a system work out a new rule, rather than recognise a familiar answer?
Key information
- Focus
- Novel problem solving and learning efficiency in the ARC-AGI benchmark family.
- Measure
- Version-specific task success; newer interactive tasks also account for efficiency.
- Version / scope
- ARC-AGI-1, ARC-AGI-2 and ARC-AGI-3 are distinct
- Update pattern
- Benchmark and competition releases, with separate verified and community submissions.
- Publisher update
- Not consistently disclosed
- Editorial verification
- 15 October 2026
In plain English
What makes it different?
ARC Prize maintains a family of evaluations intended to probe generalisation to unfamiliar problems. Earlier editions use transformations of small visual grids. ARC-AGI-3 introduces interactive environments. A score on one edition should not be read as a score on another.
Sources and attribution are listed below.
What it measures
Novel problem solving and learning efficiency in the ARC-AGI benchmark family.
Strengths
- Targets unfamiliar tasks instead of only recalled knowledge.
- Keeps benchmark generations and submission routes identifiable.
Limitations
- Performance in a deliberately constrained puzzle environment is not a complete measure of everyday usefulness.
- Specialised solvers, extra compute and different test splits can change results substantially.
How the evaluation works
Systems infer what a task requires from limited examples or interaction, then attempt to reach the required outcome.
The task split, allowed attempts, interaction budget and submission rules define the comparison. Verified evaluations and community submissions have different review contexts.
Configuration matters
Distinguish a bare model from a model plus search or a specialised solver. Check public versus private tasks, the ARC-AGI generation, testing policy and compute budget.
Primary methods: ARC-AGI benchmark family ↗ARC-AGI-3 ↗
Results at the publisher
Scores stay with the original source
We have not documented the permissions needed to reproduce this leaderboard here. You can still explore its methods and visit the publisher for results.
Versions & freshness
Verified 15 October 2026The official site lists ARC-AGI-3 alongside earlier editions. We do not combine their scores or infer a common update date.
Official benchmark family, ARC-AGI-3 explanation and terms accessible. Benchmark and competition releases, with separate verified and community submissions.
Our review date is not an evaluation date. Individual run dates are not consistently available; they remain unknown here. No automatic result collection is enabled.
Publication & reuse
The website terms reserve reproduction and commercial aggregation rights. No separate licence covering this directory's reproduction of leaderboard results is documented.
Open releases of particular tasks or software do not automatically cover website tables, competition submissions, logos or every ARC-AGI generation.
View the scope of the reuse decision
- Result fields
- No result fields cleared
- Public result display
- not enabled
- Storage
- not enabled
- Historical copies
- not enabled
- Updates
- not enabled
- Commercial reuse
- not enabled
- Derived ratings
- No scoped grant documented; not enabled.
- Attribution
- Credit the named publisher and link the primary sources. Attribution alone does not grant reproduction rights.
This decision does not clear mixed-source scientific recommendations. Publisher content, software, datasets and model licences are separate. No partnership is implied.
Primary sources & attribution
This guide is original scor.es commentary about ARC Prize Foundation (ARC Prize, Inc.)'s work. The links below support the identity, methods, versions and reuse review.
- Official foundation website ↗Checked 15 October 2026
- ARC-AGI benchmark family ↗Checked 15 October 2026
- ARC-AGI-3 ↗Checked 15 October 2026
- Website terms ↗Checked 15 October 2026
Editorial edition 2026-10-15.1 · reuse review directory-rights-1. Re-check due 14 November 2026.