ARC-AGI-3

ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence

Interactive, turn-based game-like environments built only from Core Knowledge priors, with no instructions: an agent must explore, infer the goal, build a world model and plan. 25 public demo environments plus 55 semi-private (API-tested) and 55 fully private (competition) environments, each verified fully solvable by untrained humans. Scoring is efficiency-based against human action baselines, so 100% means solving every level as efficiently as humans. The first interactive generation of the ARC-AGI series.

paper · website · leaderboard · dataset · code

99.9% · human 100%
GPT-6 Astra (high)
official 2026-09-03

Released: 2026-03
Maintainer: ARC Prize Foundation
Status: saturating
Contamination: low
Metric: efficiency-weighted score (percent, ↑)
Tasks: 135
Domains: reasoning tool-use
human: 100% untrained human test-takers (every environment solved by at least two of ten participants)

Notes. ARC Prize reports two harness conditions on the leaderboard: the Standard harness (model carries forward its own notes; provider-neutral) and the Provider Adapter harness (preserves opaque reasoning state and compacts context). Frontier models scored below 1% at the March 2026 launch; GPT-6 Astra reached 99.9% under the Provider Adapter harness in September 2026, so status is 'saturating' even though the Standard-harness score is 62.7%.

Full ledger

System	Developer	Score	Date	Source	Conditions
GPT-6 Astra (high)	OpenAI	99.9%	2026-09-03	official	split: semi-private scaffold: Provider Adapter reasoning_effort: high cost_usd_per_task: 342.13 ARC Prize verified run; $18,817 total across 55 semi-private environments (cost per environment shown). Best Provider Adapter result; verified table shows 99.95%.
GPT-6 Astra (max)	OpenAI	62.7%	2026-09-03	official	split: semi-private scaffold: Standard harness reasoning_effort: max cost_usd_per_task: 474.51 ARC Prize verified run; $26,098 total across 55 semi-private environments (cost per environment shown). Best Standard-harness result.
Opus 4.6 (Max)	Anthropic	0.5%	2026-03-24	paper	split: semi-private reasoning_effort: max Table 2 of the paper: semi-private leaderboard scores for frontier models at release.
Gemini 3.1 Pro Preview	Google DeepMind	0.4%	2026-03-24	paper	split: semi-private Table 2 of the paper: semi-private leaderboard scores at release.
GPT 5.4 (High)	OpenAI	0.2%	2026-03-24	paper	split: semi-private reasoning_effort: high Table 2 of the paper: semi-private leaderboard scores at release.