Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

Benchmark Model Rank Results
reading-comprehension-on-raceNemotron 3 Nano 30B-A3B Base#6Accuracy: 88.04
agents-on-bfclv4NVIDIA-Nemotron-3-Nano-30B-A3B-BF16#12Accuracy: 53.8
instruction-following-on-ifbenchNVIDIA-Nemotron-3-Nano-30B-A3B-BF16#17Accuracy: 71.5
math-word-problem-solving-on-mathNemotron 3 Nano 30B-A3B Base#17Accuracy: 82.88
code-generation-on-scicodeNVIDIA-Nemotron-3-Nano-30B-A3B-BF16#20Accuracy: 33.3
common-sense-reasoning-on-winograndeNemotron 3 Nano 30B-A3B Base#32Accuracy: 79.64
code-generation-on-livecodebench-v6NVIDIA-Nemotron-3-Nano-30B-A3B-BF16#36Pass@1: 68.3
arithmetic-reasoning-on-gsm8kNemotron 3 Nano 30B-A3B Base#47Accuracy: 92.34
multi-task-language-understanding-on-mmlu-proNVIDIA-Nemotron-3-Nano-30B-A3B-BF16#47Accuracy: 78.3
code-generation-on-humanevalNemotron 3 Nano 30B-A3B Base#70Pass@1: 78.05