ARC Prize 2024

The Grand Prize remains unclaimed.

The official 2024 Technical Report.

2024 competition results
Top solution approaches
In-depth technical analysis
Trends & key insights

Read the Report

All scores & papers below are open source & reproducible.

2024 High Score Winners

1st	the ARChitects Code Paper	53.5%	$25k
2nd	G. Barbadillo Code Paper	40%	$10k
3rd	alijs Code	40%	$5k
4th	William Wu Code	37%	$5k
5th	PoohAI Code Paper	37%	$5k

Score (Private Evaluation)

2024 Paper Award Winners

1st Place - $50k

"Combining Induction and Transduction for Abstract Reasoning".

Li et al.

2nd Place - $20k

"The Surprising Effectiveness of Test-Time Training for Abstract Reasoning".

Akyürek et al.

3rd Place - $5k

"Searching Latent Program Spaces".

Bonnet & Macfarlane

Runners Up - $2.5k

"The LLM ARChitect: Solving ARC-AGI Is a Matter of Perspective".

Franzen et al.

"Omni-ARC".

Barbadillo

"Mini-ARC: Solving Abstraction and Reasoning Puzzles with Small Transformer Models".

Fletcher-Hill

"Towards Efficient Neurally-Guided Program Induction for ARC-AGI".

Ouellette

"A 2D nGPT Model For ARC Prize".

Puget

2024 ARC-AGI-Pub High Scores

o3 (coming soon)	75.7%	82.8%
Jeremy Berman	53.6%	58.5%	Code Paper
MARA(BARC) + MIT	47.5%	62.8%	Code Paper
Ryan Greenblatt	43%	42%	Code Paper
o1-preview	18%	21%	Code
Claude 3.5 Sonnet	14%	21%	Code
GPT-4o	5%	9%	Code
Gemini 1.5	4.5%	8%	Code

Score (semi-private eval)

/ Score (public eval)

ARC Prize 2025

We aspire to grow ARC Prize from its experimental origins into a durable north star for AGI.

The 2025 edition of the competition will account for a diversity of incentives to serve academics, independent researchers, startups, and big labs.

Alongside the competition launch, expect to see ARC-AGI-2 - same format, better benchmark.

We'll announce more competition details early next year. Stay tuned!