4B (dense/hybrid SLM)
NVIDIA Nemotron Open Model License
Open-weight small language model release in the Nemotron 3 family with BF16, FP8, and GGUF variants for flexible deployment. Official model materials describe Nano 4B as a unified reasoning/non-reasoning model with controllable reasoning behavior (can emit intermediate reasoning traces or produce direct final answers), targeting local assistants, coding support, and edge or application-level automation in English. The model is described as compressed from NVIDIA-Nemotron-Nano-9B-v2 using the Nemotron Elastic framework, with a hybrid architecture primarily composed of Mamba-2 and MLP layers plus a small number of attention layers. Card benchmarks include reasoning-off and reasoning-on evaluations across tool-use, instruction-following, reasoning, and long-context tasks, with published references to BFCL/IFEval/IFBench/Tau2 and AIME25/MATH500/GPQA/LCB style evaluations. Release assets include open model checkpoints plus linked pre-training and post-training dataset collections in the Nemotron ecosystem.