Knowledge Bases and Evaluation Organizations: ARC Prize Foundation
Knowledge Bases and Evaluation Organizations · ARC Prize Foundation. ARC-AGI abstract reasoning benchmark and verified leaderboard. This source has 8 entries admitted under this category's review policy; each links directly to the original.
Authority tiers in this category: A2–B. Admission policy: All core resources are included. Each secondary source is admitted according to its own database, methodology, versions and leaderboard structure; there is no cross-organization scoring scheme.
ARC-AGI-3: an interactive reasoning benchmark
Tests agents' ability to explore, build world models and plan efficiently in unfamiliar interactive environments.
Evidence use: Supports ARC-AGI-3's task definition, design goals and public baselines.
Limitations: Scores on public environments may be affected by targeted optimization. Does not represent general intelligence as a whole.
Related concepts: AI Agent · Reasoning Models · Model Evaluation and Benchmarks
Link last checked: 2026-09-24
ARC-AGI-3 Technical Report
Describes the ARC-AGI-3 environments, interaction protocol, scoring, official leaderboard and anti-overfitting policy.
Evidence use: Supports precise interpretation of the benchmark's methodology, scoring rules and verification policy.
Limitations: The technical report corresponds to a specific version. Later leaderboard policies may change.
Related concepts: AI Agent · Model Evaluation and Benchmarks
Link last checked: 2026-09-24
ARC-AGI Official Leaderboard
Shows model scores, costs, dates and verification status for ARC-AGI-1, 2 and 3 in one place.
Evidence use: Supports comparisons of ARC-AGI scores tied to a date and testing policy.
Limitations: Preview and self-reported community results are not equivalent to official verification. Rankings update with new tests.
Related concepts: Reasoning Models · Model Evaluation and Benchmarks
Link last checked: 2026-09-24
ARC-AGI-1
Tests induction of novel abstract transformation rules from a few input–output examples.
Evidence use: Supports describing ARC-AGI's original task definition and public evaluation.
Limitations: Scores must be tied to the test set, cost and verification status. A single abstract reasoning benchmark does not represent general intelligence as a whole.
Related concepts: Reasoning Models · Model Evaluation and Benchmarks
Link last checked: 2026-09-24
ARC-AGI-2
Raises the demands on compositional reasoning, symbolic interpretation and generalization through redesigned tasks.
Evidence use: Supports explaining ARC-AGI-2's design changes from the first generation.
Limitations: Scores must be tied to the test set, cost and verification status. A single abstract reasoning benchmark does not represent general intelligence as a whole.
Related concepts: Reasoning Models · Model Evaluation and Benchmarks
Link last checked: 2026-09-24
ARC Prize 2024 Technical Report
Summarizes the first competition, entrants' methods, verified results and findings for abstract reasoning research.
Evidence use: Supports citing the methods and formal conclusions of the 2024 competition.
Limitations: Scores must be tied to the test set, cost and verification status. A single abstract reasoning benchmark does not represent general intelligence as a whole.
Related concepts: Reasoning Models
Link last checked: 2026-09-24
ARC-AGI-2 Technical Report
Describes the second generation's task design, difficulty control, data construction and baseline performance.
Evidence use: Supports precise explanation of how ARC-AGI-2 is constructed and scored.
Limitations: Scores must be tied to the test set, cost and verification status. A single abstract reasoning benchmark does not represent general intelligence as a whole.
Related concepts: Reasoning Models · Model Evaluation and Benchmarks
Link last checked: 2026-09-24
ARC-AGI Testing Policy
Specifies the private test, submission verification, data access and leaderboard eligibility requirements.
Evidence use: Supports judging the formality and comparability of ARC-AGI scores.
Limitations: Scores must be tied to the test set, cost and verification status. A single abstract reasoning benchmark does not represent general intelligence as a whole.
Related concepts: Model Evaluation and Benchmarks
Link last checked: 2026-09-24