NVIDIA Releases Nemotron-Cascade 2: An Open 30B MoE with 3B Active Parameters, Delivering Better Reasoning and Strong Agentic Capabilities

Nemotron-Cascade 2 was released by NVIDIA, nVIDIA announced Nemotron-Cascade 2 as an open-weight 30B Mixture-of-Experts model with 3B activated parameters.

Has 3B activated parameters, nemotron-Cascade 2 is reported to have 3 billion activated parameters within a 30 billion parameter Mixture-of-Experts architecture.

Achieved Gold Medal-level performance on IMO, IOI, and ICPC World Finals, the model achieved Gold Medal-level performance in IMO, IOI, and ICPC World Finals according to the release.

Reported benchmark scores on multiple reasoning and coding benchmarks, reported benchmark scores versus Qwen3.5-35B-A3B and Nemotron-3-Super-120B-A12B include AIME 2025 (92.4), HMMT Feb25 (94.6), LiveCodeBench v6 (87.2), IOI 2025 (439.28), ArenaHard v2 (83.5), and IFBench (82.9).

Training pipeline used SFT, Cascade RL, and MOPD, the post-training pipeline started from Nemotron-3-Nano-30B-A3B-Base and included supervised fine-tuning, Cascade RL, and MOPD.