ARC-AGI 3 controversy: Kaggle limits prevent frontier model API evaluation

JFPuget · x · 2026-08-31

JF Puget criticized the ARC-AGI 3 setup, noting that Kaggle's limited compute prevented running frontier models (unlike last year's ARC-AGI 2), making private scores impossible. François Chollet clarified that the Kaggle competition is strictly for the private set, while the separate evaluation pipeline for frontier model APIs (run upon partner request) still exists but was not utilized for this public competition phase.

Related event: Chollet responds to ARC-AGI benchmark scoring dispute(4 posts)→

Original post →

More from Companies & People

Companies & People channel →