Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models

This paper proposes Nemotron-Cascade, a framework for developing general-purpose reasoning models using cascaded domain-wise reinforcement learning (Cascade RL). It addresses heterogeneity in RL infrastructure by orchestrating sequential, domain-wise RL, achieving state-of-the-art performance across competitive programming, math, and software engineering benchmarks, and enabling models to operate in both "instruct" and "deep thinking" modes.