ARC-AGI-2 Competition

ARC Prize 2026

The ARC-AGI-2 track challenges participants to build systems that solve static reasoning tasks from the ARC-AGI-2 benchmark.

Your objective: Reach 85% accuracy on the ARC-AGI-2 private evaluation dataset within the Kaggle efficiency limits.

For general rules and the spirit of ARC Prize, see the ARC Prize 2026 overview.

Get Started

New to ARC-AGI-2? The technical guide covers data structure, development tools, and solution approaches.

Read the ARC-AGI-1 & 2 Guide

Prizes

TOTAL PRIZES AVAILABLE: $700,000

  • Progress Prizes: $275,000
  • Grand Prize: $275,000
  • Bonus Prize: $150,000

In line with the spirit of the competition, participants eligible for a prize will be removed from the competition if they do not open source their solutions.

ARC-AGI-2 Progress Prizes: $275,000

  • First Prize: $75,000
  • Second Prize: $50,000
  • Third Prize: $40,000
  • Fourth Prize: $35,000
  • Fifth Prize: $25,000
  • Sixth Prize: $20,000
  • Seventh Prize: $15,000
  • Eighth Prize: $15,000

ARC-AGI-2 Grand Prize: $275,000

The Grand Prize will be awarded to the highest scoring Solution Writeup based on the below criteria. All artifacts should be open sourced and attached to an official competition Solution Writeup within seven days of the competition's submission deadline to be considered eligible.

Submissions for the Grand Prize are evaluated equally across the following six criteria. Each criterion is scored on a scale from 0 (lowest) to 5 (highest), with the final score calculated as the average of all six.

Bonus Prize: $150,000

Awarded to the first eligible solution that scores at least 85% on the private evaluation set. If not met, we intend to roll this forward to 2027.

Submission Requirements

Submissions must be made through the Kaggle competition as a Kaggle notebook.

  • No internet access during evaluation
  • All code and methods must be open sourced to be eligible for prizes
  • Hardware and compute limits will be announced with the competition launch

Scoring Methodology

For each task, you should predict exactly 2 outputs for every test input grid. If any of the 2 predicted outputs matches the ground truth exactly, you score 1 for that task, otherwise 0. The final score is the average across all task test outputs.

Previous Years

The ARC-AGI-2 format was introduced in 2025, building on the original ARC-AGI-1 format used from 2020-2024.