📌 About the Events

The ARC Prize 2026 (ARC-AGI-2) is a premier machine learning benchmark hosted on Kaggle and co-founded by François Chollet (creator of Keras) and Mike Knoop. Unlike traditional AI leaderboards that measure memorization or LLM pattern recognition over massive text corpora, the Abstraction and Reasoning Corpus (ARC) measures an AI’s ability to acquire entirely new skills on the fly. It serves as a true test of Artificial General Intelligence (AGI), presenting systems with abstract, visual grid-based puzzles that they have never seen during training, forcing them to reason exactly like a human.

---

📍 Event Overview

• Category: Artificial General Intelligence (AGI) & Logic Reasoning Hackathon.
• Organizer: ARC Prize
• Venue: Online / Virtual via Kaggle Notebooks.
• Team Size: 1 – 5 Members (Team mergers allowed until deadline).
• Cost: Free.

---

📅 Important Dates

• Competition Kickoff: March 25, 2026
• Milestone 1 Deadline: June 30, 2026
• Milestone 2 Deadline: September 30, 2026
• Entry / Team Merger Deadline: October 26, 2026
• Final Code Submission Deadline: November 2, 2026
• Winners Announcement: December 4, 2026
• Status: Active Development Phase.

---

💰 Prizes & Incentives

• Total Prize Pool: $700,000 across multiple performance milestones.
• 🥇 Progress Prizes: $275,000 (Distributed among the top 8 teams on the final leaderboard, with 1st place securing $75,000).
• 🏆 Grand Prize Writeup: $275,000 (Awarded to the team providing the highest-scoring solution write-up explaining why their architecture works).
• 🎯 Accuracy Bonus: $150,000 (Unlocked and split among the top 5 teams that cross the legendary threshold of $\ge 85\%$ accuracy on the private evaluation set).

---

🔥 Why Participate?

• AGI Frontier: Work on the specific benchmark that standard LLMs fail at. This is the absolute cutting edge of non-memorized AI reasoning.
• Compute Provided: Kaggle supplies specialized execution sandboxes (up to 12 hours of dual-use CPU/GPU processing) to keep the playing field structurally level.
• Massive Career Signal: Ranking highly on an ARC leaderboard is a premier credential recognized by elite research labs (OpenAI, Anthropic, Google DeepMind).
• Open Source Contribution: All prize-winning code must be made fully public, meaning your breakthroughs will directly shape the future of machine learning.

---

⚖️ Rules & Eligibility

Evaluation Metric: Systems are given a grid matrix task and must output exactly 2 prediction attempts. If either matches the hidden ground truth grid layout exactly, you score 1, otherwise 0.

Who: Open globally to all data scientists, ML engineers, algorithmic researchers, and students.

Submission Format: Notebook submission via Kaggle. External data and pre-trained models are allowed, provided they are freely and publicly available.

Strict Constraints: No internet access is allowed during the automated container rerun evaluation. Your submission must generate its predictions entirely locally within 12 hours.