activeReasoning

ARC-AGI-3 — Standard harness

Metric: Score (higher is better)Introduced: 2026

ARC-AGI-3 is the ARC Prize Foundation's interactive benchmark of novel turn-based game environments with no instructions, rules, or stated goals. Scoring compares how efficiently a model completes each level versus human players. The Standard harness is the Foundation's consistent cross-provider comparison and must not be merged with the Provider Adapter harness or with ARC-AGI-2 (arc-agi-2).

Models ranked

1

tracked on this benchmark

Score band

79.3 – 79.3

best → lowest tracked

Snapshot trend

—

need ≥2 snapshots

Leaderboard

Tracked models ranked by Score (higher is better).

Compare candidates
#Model variant and provenanceScore
1
Claude Fable 5.2

Configuration: claude-fable-5-2 (High, Standard harness)

Version: ARC-AGI-3 Standard harnessHarness: ARC Prize Standard harnessEvaluator: ARC Prize FoundationObserved: Nov 9, 2026Confidence: self-reportedSource

Notes: ARC Prize Foundation figures as reported in the Claude Fable 5.2 system card §8.15 (retrieved 2026-11-09): verified score 79.26% with the Standard harness at high effort (highest Standard-harness score published there). Other Foundation figures in the same section: Provider Adapter 97.72% at high; Standard 65.66% and Provider Adapter 94.90% at max. This is not ARC-AGI-2 and must not be written onto arc-agi-2. No separate arc-agi-3 file was found in the ARC v2.json dump (that file is ARC-AGI-2 semi-private). Owner attribution is the Foundation as quoted in the system card; treat as Foundation-reported rather than a live board scrape.

79.3

How to read this benchmark

This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.

Trust this score when

  • There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
  • The model list covers the same version family you can actually deploy today.
  • Top candidates overlap with your required routing and feature requirements.

Be cautious when

  • There is only one benchmark snapshot or the dataset appears stale.
  • The benchmark metric direction is opposite of your decision objective.
  • The score difference between options is narrow and likely within implementation variance.

Related benchmarks

ANLS*
Average Normalized Levenshtein Similarity*
ARC
AI2 Reasoning Challenge
ARC-AGI-2
ARC-AGI-2 High
ARC-AGI-2 — High Effort
ARC-E
AI2 Reasoning Challenge-Easy
ARC Prize / ARC Challenge

Last reviewed: Nov 9, 2026

Resources

Website