NVIDIA Nemotron 3 Super 120B-A12B
Release in the NVIDIA Nemotron family · version 3 Super (120B-A12B), v1.0
Nemotron 3 Super is an NVIDIA language model with 120B total and 12B active parameters, using a hybrid Mamba-2 and attention architecture with latent mixture-of-experts layers and multi-token prediction, and configurable reasoning. NVIDIA released BF16, FP8, and NVFP4 checkpoints and a separate base checkpoint on Hugging Face.15
- Model hub: Model card (BF16) (external site: huggingface.co)
- License: NVIDIA Nemotron Open Model License (external site: nvidia.com)
- Documentation: Nemotron 3 Super training recipe (external site: github.com)
Availability and license
Overall availability
Weights are downloadable from Hugging Face under the NVIDIA Nemotron Open Model License; the repository metadata showed no access gate when checked. The NIM container version is covered by separate NVIDIA software license terms.12
Availability is separate from permission: read the license before using or redistributing.
NVIDIA Nemotron Open Model License (external site: nvidia.com)13
Apache License 2.0 (Nemotron Developer Repository training recipes) (external site: raw.githubusercontent.com)4
The NVIDIA Nemotron Open Model License (version dated December 15, 2025) is NVIDIA's own license. It grants a perpetual, worldwide, royalty-free license to reproduce, modify, and distribute the model and derivatives, and states that the works are commercially usable and that NVIDIA claims no ownership of outputs. Redistributors must pass on a copy of the license and keep attribution notices, including a specified NVIDIA attribution statement when a NOTICE file is present. The license ends for anyone who brings patent or copyright litigation claiming the work or its outputs infringe. It also requires users to indemnify NVIDIA against third-party claims arising from their use or distribution, requires compliance with export and sanctions laws, and is governed by U.S. and Delaware law.3
Model-disclosure tier
Open-weight, plus published inference code, training code, and training recipe, and at least documented training-data composition.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | BF16, FP8, and NVFP4 post-trained checkpoints and a BF16 base checkpoint are published on Hugging Face.152 |
| Inference codeIs code for running the model published? | Public | The model card gives deployment instructions for vLLM, SGLang, TensorRT-LLM, and Transformers.1 |
| Training codeIs the code used to train the model published? | Public | The Apache-2.0 Nemotron Developer Repository provides pretraining, SFT, RL, and evaluation stages built on Megatron-Bridge, NeMo RL, and NeMo Evaluator.54 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The model card lists the pre- and post-training datasets. Major portions are released in Hugging Face collections, some requiring access approval, while several third-party and NVIDIA datasets are listed as private and not publicly accessible.16 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Public | The recipe documents each stage described in the model card (pretraining, SFT, and RL) plus quantization and evaluation. NVIDIA notes the recipes use only the open-sourced data subset, so results will differ, and marks a knowledge-distillation stage as not yet published.51 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Public | The model card publishes benchmark results collected with the NeMo Evaluator SDK; it says some benchmarks used internal scaffolding not yet open-sourced.1 |
What it is useful for
The model card lists agentic workflows, long-context reasoning, high-volume workloads such as IT ticket automation, tool use, and retrieval-augmented generation, and states the model is ready for commercial use.1
Run and use notes
- The model card documents serving with vLLM, SGLang, TensorRT-LLM, and Transformers, and states a minimum of 8x H100-80GB GPUs for the BF16 checkpoint; it points to the NVFP4 checkpoint for running on a single B200 or DGX Spark.1
Organization context
Provenance and derivatives
Post-trained by NVIDIA (supervised fine-tuning and reinforcement learning) from its own pre-trained Nemotron 3 Super base checkpoint. The model card says post-training synthetic data was generated with teacher models, including third-party open models such as GPT-OSS-120B.1
- Derived from: NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16 (external site: huggingface.co) — NVIDIA's pre-trained base checkpoint for this release.
Other releases in the NVIDIA Nemotron family
- NVIDIA Nemotron 3 Ultra 550B-A55BModel-disclosure tier (USASI rubric v0.1): Open-weight
- NVIDIA Nemotron 3.5 Lightning 30B-A3BModel-disclosure tier (USASI rubric v0.1): Open-stack
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.