Connect Four

The 'Connect Four' dataset/benchmark is used to evaluate the capability of AI systems to autonomously implement end-to-end machine learning pipelines for the game Connect Four, specifically through a proof-of-concept involving AlphaZero-style algorithms.

Papers using Connect Four (2)

AlphaZero in Sparsely Rewarded Games: Limits and Auxiliary Supervision2026Frontier Coding Agents Can Now Implement An Alphazero Self-play Machine Learning Pipeline For Connect Four That Performs Comparably To An External Solver2026