NVIDIA Nemotron 3.5 Super VL 120B-A12B
NVIDIA Nemotron 3.5 Super VL 120B-A12B is an open-weight model release in the NVIDIA Nemotron family from NVIDIA1, version 3.5 Super VL (120B-A12B), v1.0 GA, released Oct 15, 2026. Its weights can be downloaded by the public125 under the OpenMDW License Agreement, version 1.113.
Nemotron 3.5 Super VL is an NVIDIA vision-language model with 120B total and 12B active parameters. It pairs a hybrid Mamba-2 and attention mixture-of-experts language backbone with a C-RADIO vision encoder, accepts text, image, and video input with text output, supports a context of up to 1M tokens, and includes a multi-token prediction head for speculative decoding. NVIDIA published BF16 and NVFP4 post-trained checkpoints and a text-only base checkpoint.156
- Model hub: Model card (BF16) (external site: huggingface.co)
- License: OpenMDW License Agreement, version 1.1 (external site: openmdw.ai)
- Documentation: Nemotron 3.5 Super training recipe (external site: github.com)
- Release notes: Release announcement (external site: blogs.nvidia.com)
On AI on Hugging Face: nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16
Availability and license
Overall availability
Weights are downloadable from Hugging Face under the OpenMDW-1.1 license named in the model card; the repository metadata showed the repository as public with no access gate when checked.12
Availability is separate from permission: read the license before using or redistributing.
OpenMDW License Agreement, version 1.1 (external site: openmdw.ai)13
Plain-language guide to the OpenMDW License Agreement, version 1.1Apache License 2.0 (Nemotron Developer Repository training recipes) (external site: raw.githubusercontent.com)4
Plain-language guide to the Apache License 2.0
The model card's license field and its terms-of-use section name OpenMDW-1.1 and link to the text on openmdw.ai; no separate LICENSE file was present in the model repository when checked. OpenMDW-1.1 grants free permission to use, modify, and share the model materials without restriction, including under copyright, patent, database, and trade secret rights. Redistributors must include a copy of the agreement and keep applicable notices of origin. Rights end for anyone who sues claiming the materials infringe a patent or copyright, unless responding to a suit brought first. It places no conditions on outputs, disclaims warranties, and leaves clearing third-party rights to the user. No SPDX ID is recorded for version 1.1.123
Component reuse rights
- weights
- Unknown — no complete fact-level rights review
- code
- Reviewed qualifying license recorded — check scope and conditions
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
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 | The BF16 post-trained weights are published on Hugging Face; NVIDIA's recipe also lists NVFP4 and text-only base BF16 checkpoints.125 |
| Inference codeIs code for running the model published? | Public | The model card gives serving instructions for vLLM and SGLang; the repository includes custom Transformers modeling code.12 |
| Training codeIs the code used to train the model published? | Partial | The Apache-2.0 Nemotron Developer Repository provides SFT, RL, and evaluation recipes. NVIDIA states the text-only pretraining pipeline is not reproduced there, and that the agentic teacher and one vision-reasoning stage depend on NeMo Gym components that are not public.54 |
| Training-data informationDoes the information cover provenance, scope, acquisition, selection, labeling, processing, and where data or alternatives can be obtained? Access alone does not establish completeness. | Partial | The model card lists pre- and post-training datasets with collection periods, sources, and the models used to generate synthetic data, and links a public training content summary. Several third-party and NVIDIA datasets are listed as private, some with undisclosed sizes or generation models.1 |
| Training-data accessCan the training data be obtained? This is independent of information completeness and reuse rights; original unshareable data need not be downloadable. | Partial | The pretraining corpus is drawn from NVIDIA's public Nemotron pretraining dataset collection, while the SFT and RL data used for this release are not yet published and some listed datasets are private.15 |
| Complete training pipelineIs the complete base-training and preprocessing pipeline published, including configuration? Fine-tuning code or an inference SDK alone is insufficient. | Partial | Post-training stages are published as reference recipes; the base pretraining pipeline is not, because it depends on teacher logits and data that NVIDIA says are not public.5 |
| Legacy data assessment (v0.1)Historical assessment combining download access and disclosure. Preserved for traceability; excluded from the v0.2 tier calculation. See the new separate assessments above. | Unknown | Not assessed. |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The model card and recipe describe each training stage; NVIDIA links a technical report. The recipe says the SFT and RL data are not yet published and that MTP boosting is not included, so recipes run on the user's own data.15 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The model card publishes benchmark results collected via NeMo Gym and points to its reproducibility recipes; it says several vision benchmarks used official implementations or internal scaffolding not yet open-sourced.1 |
What it is useful for
The model card lists reasoning, coding, tool use, document understanding, and video understanding, and describes the BF16 weights as intended for customization, domain adaptation, quantized variants, and full-precision evaluation. It states the model is ready for commercial or non-commercial use.1
Run and use notes
- The model card documents serving the BF16 checkpoint with vLLM or SGLang on 4x H100 (tensor parallel 4) or 2x GB200 (tensor parallel 2), with native MTP speculative decoding, and points to the NVFP4 checkpoint for DGX Spark or more efficient data-center inference. It lists Linux as the operating system and NVIDIA Hopper and Blackwell as supported hardware.1
Organization context
Provenance and derivatives
NVIDIA initialized the language backbone from its Nemotron 3 Super Base model, distilled it from Nemotron 3 Ultra Base, and added NVIDIA's C-RADIO vision encoder and a connector during supervised fine-tuning; the card does not state the encoder's version or lineage. The model card says synthetic training data was generated or filtered with many third-party models, including non-U.S. models such as Qwen, DeepSeek, Kimi, and GLM releases, alongside U.S. models such as gpt-oss-120b and Gemma. These were data-generation tools, not base weights.15
- Derived from: Nemotron 3 Super Base — NVIDIA base model used to initialize the language backbone.
- Derived from: C-RADIO vision encoder — NVIDIA vision encoder named in the model card; version not stated.
Other releases in the NVIDIA Nemotron family
- NVIDIA Nemotron 3 Super 120B-A12BModel-disclosure tier (USASI rubric v0.2): Open-weight
- NVIDIA Nemotron 3 Ultra 550B-A55BModel-disclosure tier (USASI rubric v0.2): Open-weight
- NVIDIA Nemotron 3.5 Lightning 30B-A3BModel-disclosure tier (USASI rubric v0.2): Open-weight
- NVIDIA Nemotron Labs Dory1.0-4BModel-disclosure tier (USASI rubric v0.2): Open-weight
What this catalog does not know
- Legacy data assessment (v0.1): unknown.
Have a primary source? How to report a correction.
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.
Share or cite this page
- X (opens X)
- Bluesky (opens Bluesky)
- LinkedIn (opens LinkedIn)
- Reddit (opens Reddit)
- Hacker News (opens Hacker News)
- Email (opens your email app)
- RSS updates
Cite this page
APA style
USASI. (2026, October 18). NVIDIA Nemotron 3.5 Super VL 120B-A12B. United States of America Superintelligence. https://unitedstatesofamericasuperintelligence.com/open/nemotron-3-5-super-vl-120b-a12b/
BibTeX
@misc{usasi_nemotron_3_5_super_vl_120b_a12b,
author = {{USASI}},
title = {NVIDIA Nemotron 3.5 Super VL 120B-A12B},
year = {2026},
month = oct,
howpublished = {\url{https://unitedstatesofamericasuperintelligence.com/open/nemotron-3-5-super-vl-120b-a12b/}},
note = {United States of America Superintelligence. Last updated 2026-10-18}
}The date is when this page was last updated. Add the date you read it if your style needs one.
For maintainers: a “Listed in USASI” badge
Links to this page. It says the entry is listed, not endorsed or rated. About the badge
Markdown (for a README)
[](https://unitedstatesofamericasuperintelligence.com/open/nemotron-3-5-super-vl-120b-a12b/)
HTML
<a href="https://unitedstatesofamericasuperintelligence.com/open/nemotron-3-5-super-vl-120b-a12b/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/listed.svg" alt="Listed in the USASI catalog" height="20"></a>