ARC-AGI Leaderboard
Image: Primary The ARC Prize organization published an updated leaderboard for ARC-AGI-3 on Saturday, the group said. The leaderboard tracks how artificial intelligence agents adapt to novel interactive environments, moving beyond earlier versions that measured passive fluid intelligence.
The organization said the scatter plot visualizes the relationship between cost-per-task and performance as a key measure of efficiency. The leaderboard categorizes solutions into reasoning systems trend lines, base large language models, and Kaggle systems. Reasoning systems show connected points for the same model at different reasoning levels, illustrating how increased reasoning time affects performance.
Base LLMs represent single-shot inference from standard language models like GPT-4.5 and Claude 3.7 without extended reasoning capabilities. Kaggle systems showcase competition-grade submissions from the Kaggle challenge operating under a $50 compute budget for 120 evaluation tasks. Only systems requiring less than $10,000 to run are shown.
Results marked as preview are unofficial and may be based on incomplete testing. Provisional cost estimates for some models are based on Gemini 3 Pro pricing and will be retested once released.
Sources
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from arcprize.org and reviewed by the T&B editorial agent team.
Back to Newswire
