Research institute · Research orgs · San Francisco, California · Est. 2024
ARC Prize Foundation is a U.S.-based nonprofit organization that accelerates progress toward Artificial General Intelligence (AGI) by creating, curating, and maintaining human-calibrated benchmarks centered on fluid intelligence and generalization. Its core mission is to identify and measure the “capability gap” between human and artificial intelligence on tasks that are straightforward for people but challenging for contemporary AI systems, and then to close that gap by guiding researchers toward approaches beyond pattern matching and memorization.
Operationally, the organization builds a benchmark suite (ARC-AGI-1/2/3) and wraps it in an evaluation and verification ecosystem: open-source datasets and benchmarking tools, a verified leaderboard intended to certify model performance using official hidden evaluation data, and an accompanying “ARC Prize” competition that provides monetary incentives for open-source solutions. ARC-AGI-3 extends the program to interactive, agentic settings (browser/API play or SDK-based evaluation) where agents must explore, infer goals, model environment dynamics, and plan actions across long horizons without explicit instructions. Strategically, ARC Prize Foundation positions itself as an “independent standard” for measuring progress: it publishes a testing/verification policy, discloses donors on its donation page, and states that sponsors/donors do not influence benchmark roadmaps, scoring, or verification eligibility. It also publishes analysis and auditing-oriented materials (e.g., human-performance measurement and verification/audit program announcements) to help interpret benchmark results and to reduce the risk of misleading conclusions about frontier systems.
No indexed openings right now. Check the careers page ↗.
OpenAI's GPT-6 Astra reaches state-of-the-art results on ARC-AGI-3
ARC Prize 2026: ARC-AGI-3 Milestone Prize #1Celebrating the top three submissions from our first ARC-AGI-3 milestone prize.
Analyzing GPT-5.5 & Opus 4.7 with ARC-AGI-3Analyzing GPT-5.5 & Opus 4.7 with ARC-AGI-3 AI benchmarks can be incredible tools, but they usually only tell you if a model passed or failed. With ARC-AGI-3, however, we can see the thought process *behind* the score, not just the outcome. This week we went through 160 replays a
Measuring Human Performance on ARC-AGI-3Measuring Human Performance on ARC-AGI-3 AGI is here when a system can learn like a human. However there is still a gap between what humans can learn and what AI can learn. ARC Prize Foundation exists to understand this gap. The ARC-AGI benchmarks are our tools for measuring it.
Announcing ARC-AGI-3ARC-AGI-3 is the first fully-interactive benchmark in the ARC-AGI series, featuring hundreds of handcrafted games with thousands of levels.
ARC Prize 2025 Results and AnalysisWinners, analysis, and interviews.
Announcing ARC Prize VerifiedARC-AGI benchmark test results you can trust.
ARC-AGI-3 Preview: 30-Day LearningsOne Month of Learnings Building Interactive Reasoning Benchmarks
The Hidden Drivers of HRM's Performance on ARC-AGIWe scored on hidden tasks, ran ablations, and found that performance from the Hierarchical Reasoning Model comes from an unexpected source
ARC Prize Foundation Statement on the US AI Action PlanARC Prize Response To AI Action Plan
We tested every major AI reasoning system. There is no clear winner.A technical analysis of frontier AI reasoning systems as of June 2025.
ARC-AGI-2 A New Challenge for Frontier AI Reasoning SystemsTechnical context and description of the ARC-AGI-2 Benchmark
ARC Prize Foundation reports GPT-6 Astra ARC-AGI-3 results on semi-private and describes observed behaviors in the agentic benchmark.
ARC Prize 2026: ARC-AGI-3 Milestone Prize #1ARC Prize Foundation awards its first $37.5K ARC-AGI-3 milestone prize for open-source submissions, and highlights top submissions from the ARC-AGI-3 Kaggle competition.
Analyzing GPT-5.5 & Opus 4.7 with ARC-AGI-3ARC Prize Foundation describes auditing and replay-based analysis of GPT-5.5 and Anthropic Opus 4.7 on ARC-AGI-3 and notes open-sourcing an analysis package.
Measuring Human Performance on ARC-AGI-3ARC Prize Foundation releases a human dataset for ARC-AGI-3, describing controlled study details and the purpose of measuring human performance to interpret intelligence gaps.
Announcing ARC-AGI-3ARC Prize Foundation launches ARC-AGI-3, describes the interactive agentic benchmark format, and announces ARC Prize 2026 going live with a $2M prize program and related competition structure.
ARC Prize 2025 Results and AnalysisARC Prize Foundation publishes results and analysis from ARC Prize 2025 and discusses plans for ARC-AGI-3 release in early 2026 alongside ARC Prize 2026.