ARC Prize Foundation
ARC Prize Foundation advances open-source artificial general intelligence research through human-calibrated benchmarks, competitions, prizes, and an open research community. Its central objective is to measure and close the gap between capabilities that are easy for humans and difficult for AI, especially fluid intelligence, reasoning, exploration, and adaptation.
At a Glance
- Frontier AI labs
- Academic AI and machine-learning researchers
- Open-source AI developers and competition teams
- Universities and students
- +2 more
AI Tools by ARC Prize Foundation
(1)
ARC-AGI Benchmarking
Open Source ARC-AGI Benchmark Harness
Discussions
No discussions yet
Be the first to start a discussion about ARC Prize Foundation
Latest News
OpenAI's GPT-6 Astra on ARC-AGI-3
ARC Prize 2026: ARC-AGI-3 Milestone Prize #1 awarded to the leading open-source submissions
Analysis of GPT-5.5 and Opus 4.7 with ARC-AGI-3
Release of the ARC-AGI-3 human-performance dataset from a controlled study of 458 participants
Products & Services
Human-calibrated benchmarks designed to measure fluid intelligence and generalization on tasks that are simple for people but difficult for AI; the series includes ARC-AGI-1 and ARC-AGI-2.
A harder static reasoning benchmark intended to test adaptability and efficiency beyond pattern matching; every task was solved by at least two humans in two attempts or fewer during calibration.
Interactive reasoning benchmark consisting of hundreds of handcrafted, turn-based environments and thousands of levels. Agents must explore, infer rules and goals, plan, act, and carry learning across levels.
Open-source competitions and paper prizes that reward progress on ARC-AGI; the 2026 program has ARC-AGI-2, ARC-AGI-3, and Paper Prize tracks with $2M in total prizes.
Market Position
ARC Prize positions ARC-AGI as a high-signal alternative to benchmarks dominated by memorization, factual recall, or superhuman exam performance. Its differentiation is human-calibrated tasks that are easy for people but hard for AI, plus open-source prizes and an interactive ARC-AGI-3 benchmark aimed at agentic intelligence. Comparable alternatives include general AI evaluation suites, model-provider reasoning benchmarks, and research benchmarks such as BIG-bench and Humanity's Last Exam; ARC's stated niche is fluid intelligence and efficient adaptation rather than broad task coverage.
Leadership
Founders
François Chollet
Creator of Keras, author of the 2019 paper On the Measure of Intelligence and creator of ARC-AGI; the official ARC Prize site identifies him as a co-founder, board member, and creator of ARC-AGI. He has been an AI researcher and engineer at Google.
Mike Knoop
Co-founder of Zapier, where he led engineering and product and helped drive the company's early AI work; the official site identifies him as an ARC Prize co-founder and board member. He is also a co-founder of Ndea.
Greg Kamradt
President and board member who co-led ARC Prize 2024 and became president as the initiative expanded into the foundation in 2025. Previously led operations and product at Digits and spent several years at Salesforce in strategy, growth, and predictive analytics; he also founded Leverage.
Executive Team
Greg Kamradt
President
Led ARC Prize 2024 operations and product; previously led ops and product at Digits and worked in strategy, growth, and predictive analytics at Salesforce.
François Chollet
Co-founder, Board Member, Creator of ARC-AGI
Creator of Keras and ARC-AGI, author of On the Measure of Intelligence, and former Google AI researcher/engineer.
Board of Directors
Founding Story
ARC Prize began as a 2024 competition created by Mike Knoop and François Chollet to redirect AI research toward open-source solutions to ARC-AGI. It grew out of Chollet's 2019 ARC-AGI benchmark and his hypothesis that existing approaches would struggle with fluid intelligence; after the 2024 competition and increased attention following OpenAI o3 testing, the organizers expanded it into a nonprofit foundation in 2025 to make ARC-AGI an academic, policy, and industry standard and to guide open AGI research.
Business Model
Revenue Model
A donation- and sponsorship-funded nonprofit model. ARC Prize publicly solicits tax-deductible donations and uses sponsor contributions to fund benchmark development, infrastructure, prizes, research, and community programs; the organization also works with competition and research partners.
Target Markets
- Frontier AI labs
- Academic AI and machine-learning researchers
- Open-source AI developers and competition teams
- Universities and students
- AI standards organizations and regulators
- Technology companies evaluating reasoning and agentic systems
- AI research on abstraction, reasoning, program synthesis, and efficient learning
- Evaluating frontier foundation models and reasoning systems
- Testing autonomous agents' exploration, planning, and adaptation
- Benchmarking model progress for AI labs, academic researchers, and developers
- Open-source competition and reproducible research toward AGI
- AI standards and policy evaluation
- OpenAI
- Anthropic
- Google DeepMind
- xAI