Code Arena is a public AI coding model evaluation platform operated by Arena (formerly LMArena), launched on November 12/17, 2025. Rather than testing isolated code snippets, it measures models' ability to agentically plan, scaffold, and iteratively build complete web applications. Users submit a prompt, multiple models generate live, deployable apps that are then compared in blind pairwise voting and aggregated in a real-time public leaderboard. Every action is fully logged, restorable, and scientifically reproducible.

Zdroje (3)