The ARC Prize 2026 - Paper Track invites participants to document their conceptual approach for the ARC Prize 2026. This track is designed for those who have participated in either the ARC-AGI-2 or ARC-AGI-3 prediction competitions and wish to submit a Writeup detailing their solution.
Real intelligence is defined not by memorizing answers, but by knowing what to do when problems change. The ARC Prize focuses on true generalization: whether a system can quickly learn new skills in unfamiliar situations. Unlike benchmarks that reward pattern recognition on known tasks, this competition evaluates how well systems adapt to new problems they have never encountered before. The evaluation environment prevents simple memorization by utilizing hidden, interactive environments that require exploration and multi-step reasoning.
In this track, participants develop approaches that learn quickly, generalize well, and solve problems never seen before. To participate, you must submit a Writeup documenting your submission for either ARC-AGI-2 or ARC-AGI-3. Your solution aims to help move AI closer to systems that learn like people: flexible, efficient, and capable of handling new challenges.
Submission Requirements A valid submission must include: * Kaggle Writeup: Serves as the project report, including a title, subtitle, and detailed analysis. It must not exceed 1,500 words. * Media Gallery: A required cover image and any associated images or videos. * Public Notebook: Code submitted as a publicly accessible notebook (no login or paywall required). * (Optional) Public Project Link: An alternative to the Kaggle Writeup where a PDF version of the paper can be uploaded.
Evaluation Criteria Submissions are evaluated equally across six criteria, each scored from 0 to 5, with the final score being the average: 1. Accuracy: Performance on the leaderboard. 2. Universality: How general and universal the approach is beyond the competition. 3. Progress: How much the paper increases the chance of achieving a top score on the ARC Prize. 4. Theory: Description of why the submission works, rather than just how. 5. Completeness: Thoroughness of the coverage of the submission. 6. Novelty: Novelty relative to existing public research.
Prizes The Paper Award is given to three submissions with the most points. Additionally, papers scoring above 4.5/5 on the rubric qualify for a bonus prize pool of $375,000, which is divided equally among all qualifying papers.