Paper Prize
ARC Prize 2026
The Paper Prize rewards conceptual progress that best advances our understanding of how to achieve strong performance on ARC-AGI.
Paper submissions must be linked to a Kaggle code submission (ARC-AGI-2 or ARC-AGI-3 track) that demonstrates the approach detailed in the paper. The code submission need not achieve a high score for the corresponding paper to be eligible.
For general rules and the spirit of ARC Prize, see the ARC Prize 2026 overview.
Prizes - $450K Total
Top Paper - $75K (guaranteed)
Awarded to the highest-scoring paper submissions across both competition tracks (ARC-AGI-2 and ARC-AGI-3).
- 1st Prize: $50K
- 2nd Prize: $20K
- 3rd Prize: $5K
Outstanding Papers Pool - $375K
Awarded to papers scoring above 4.5 on the paper rubric, at host discretion. Multiple parties may win from this pool.
Evaluation Rubric
Papers are evaluated equally based on the following rubric, with a score from 0 (lowest) to 5 (highest) in each category.
| Category | Description |
|---|---|
| Accuracy | How accurate is the submission based on its performance on the leaderboard? |
| Universality | How general and universal is the approach beyond the competition? Does your method generalize to other similar problems? |
| Progress | How much does the paper increase the overall chance of anyone achieving 85% on ARC-AGI? |
| Theory | How well does the paper describe why the approach works (as opposed to merely describing how it works)? |
| Completeness | How thoroughly and completely does the paper cover the submission to the leaderboard? |
| Novelty | How novel is the approach relative to existing public research? |
Each paper must include a corresponding Kaggle submission confirming it describes a real, working entry. The submission's score will be used in the rubric's “accuracy” category.
Paper rubric evaluations will not be shared. In the event of a tie, the paper entered first will be the winner.
What to Include
- Abstract: What the contribution is (e.g., “we present a method to solve ARC-AGI, with the following characteristics...”)
- Intro: What ARC-AGI is, why it's important, and the inspiration behind the approach
- Prior work: Previous approaches related to yours - highlight similarities and differences
- Approach: How it works, including an algorithm-level description
- Results: Scores on various sets (Kaggle leaderboard, public eval). Don't report train set performance
- Conclusion: Summary of the contribution and what was achieved
Shorter and clearer is always better. No filler, no unnecessary equations. Papers are about communicating ideas clearly so others can learn from and reuse them.