Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation
2026-03-19 • Computation and Language
Computation and LanguageArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors present Nemotron-Cascade 2, a large language model that is smaller but still highly capable in math and coding tasks, rivaling bigger models. It performs very well in top international competitions like the Math Olympiad and programming contests, matching results usually achieved by much larger models. They improved the model by using a refined training method called Cascade RL, combined with learning from specialized teacher models across different tasks. The authors also released the model and the training data for others to use.
MoE modelactivated parametersmathematical reasoningcoding reasoningInternational Mathematical OlympiadInternational Olympiad in InformaticsICPC World FinalsSupervised Fine-Tuning (SFT)Cascade Reinforcement Learning (Cascade RL)on-policy distillation
Authors
Zhuolin Yang, Zihan Liu, Yang Chen, Wenliang Dai, Boxin Wang, Sheng-Chieh Lin, Chankyu Lee, Yangyi Chen, Dongfu Jiang, Jiafan He, Renjie Pi, Grace Lam, Nayeon Lee, Alexander Bukharin, Mohammad Shoeybi, Bryan Catanzaro, Wei Ping
Abstract
We introduce Nemotron-Cascade 2, an open 30B MoE model with 3B activated parameters that delivers best-in-class reasoning and strong agentic capabilities. Despite its compact size, its mathematical and coding reasoning performance approaches that of frontier open models. It is the second open-weight LLM, after DeepSeekV3.2-Speciale-671B-A37B, to achieve Gold Medal-level performance in the 2025 International Mathematical Olympiad (IMO), the International Olympiad in Informatics (IOI), and the ICPC World Finals, demonstrating remarkably high intelligence density with 20x fewer parameters. In contrast to Nemotron-Cascade 1, the key technical advancements are as follows. After SFT on a meticulously curated dataset, we substantially expand Cascade RL to cover a much broader spectrum of reasoning and agentic domains. Furthermore, we introduce multi-domain on-policy distillation from the strongest intermediate teacher models for each domain throughout the Cascade RL process, allowing us to efficiently recover benchmark regressions and sustain strong performance gains along the way. We release the collection of model checkpoint and training data.