AReaL is a reinforcement learning (RL) infrastructure designed to bridge foundation model training with modern agent-based applications. It was originally developed by researchers and engineers from ...