Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning

The paper presents a method for scaling zero-reinforcement learning (RL) to a trillion parameters, achieving emergent reasoning capabilities and demonstrating competitive performance on mathematical benchmarks.