Open Model WeightsNVIDIA · FIELD VERIFIED
Primary sources last checked on 18 Oct 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.
FRESHNESS & VERIFICATION
A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.
CONFLICT INTELLIGENCE
+ Save to my shortlistMonitor changes →Local browser monitoring · no background email alerts
DIRECT ANSWER · FEATURED MODEL BRIEFING
Source-grounded editorial layer · reviewed 2026-10-18
Nemotron 3.5 Super VL BF16 is a publisher-released open-weight checkpoint in the Nemotron 3.5 Super VL family. This BF16 publisher checkpoint serves as the full-precision reference for evaluating and adapting Super VL, subject to substantial hardware and multimodal processing requirements. OMW verified its exact publisher repository revision, 65 recognized weight artifacts, openmdw-1.1 license evidence and 1,048,576 configured tokens on 18 October 2026. This source-based review does not represent an independent model inference, speed or quality benchmark.
WHY IT MATTERS
Super VL joins a publisher-declared million-token configuration with text, images and sampled video, so context cost and media preprocessing are central to real deployments.
Separate BF16 and NVFP4 repositories permit controlled precision experiments, while a 120B-total/12B-active sparse architecture still requires substantial storage.
MODEL-SELECTION VIEW
NVIDIA's BF16 release is the reference checkpoint for numerical fidelity and potential customization, but requires a deployment sized to the full weights and multimodal processing rather than merely to 12B active parameters.
The official BF16 repository contains 65 recognized artifacts. Reproduce identical screenshot and video tasks at controlled preprocessing and output length against the other Super VL format before publishing a performance comparison.
NVIDIA describes Super VL as a Mamba-2, attention and sparse-expert hybrid with a C-RADIO visual encoder. The model card states approximately 120B logical parameters with 12B active for its language processing path.
This is not a 12B stored-weight checkpoint. The visual backbone, compressed or full-precision experts, model-parallel allocations and caches all matter to the end-to-end system size.
OMW's verified configuration field holds 1,048,576 context tokens; publisher documentation includes image and video input support. Long-context limits apply at the implementation level and can be reduced substantially by serving engine settings.
For grounded visual QA, fix image dimensions and video frame sampling, then report which evidence the response refers to. Count failure cases such as missed frames, wrong page citations and tool-parser errors rather than relying on generalized capability labels.
The official BF16 checkpoint is positioned for reference inference and adaptation; NVFP4 is a separate optimized-precision artifact. Treat each weight manifest and revision as an independent test input rather than a metadata flag.
Quantized kernels, scaling data and conversion overhead can affect both memory and outputs. Compare like-for-like prompts, media processors and output constraints on matched acceleration resources.
NVIDIA's model card discusses Multi-Token Prediction as part of serving and speculative decoding. Such features need compatible engines, exact versions and possibly custom model code; they do not guarantee a universal acceleration factor.
Record the actual installed vLLM or SGLang commit, tool parser, speculative settings, CUDA kernels and media preprocessing. Separate latency of prefill and decoding and distinguish throughput per stream from concurrent service throughput.
The checked OpenMDW 1.1 declaration governs rights and conditions. It should never be summarized as MIT or Apache-2.0 merely because the weights are public and discoverable.
Our editorial recommendation is to test Super VL when visual grounding and long context are essential, then compare Bolt under the same tasks when compute footprint dominates. The point of comparison is task completion and verified cost, not different vendor leaderboard numbers.
IMPLEMENTATION & EVALUATION
Reproduction checklist for Nemotron 3.5 Super VL BF16, not an OMW inference run. Keep the exact BF16 checkpoint, tokenizer, media processing, tool harness, software revision and hardware in the report.
Record the official repository SHA, artifact manifest and applicable weight-specific license before benchmarking.
Save a reproducible revision and artifact checksum list together with the exact license source and date.
Model branches may change, and third-party conversions or processors can alter the tested input without notice.
Reproduce the model-card tokenizer, multimodal preprocessing, serving code, precision and tool-call format.
Run at least one positive and one failure-mode task with pinned environment and exact model implementation, or record incompatibility.
Declared library tags do not prove a given package version supports this architecture or compressed format.
Use representative prompt lengths, media dimensions and tool interactions while instrumenting actual memory, latency and correctness.
Retain raw input/output pairs, peak memory, prefill/decode latency, error logs and infrastructure specification.
A configured context limit, weight-only arithmetic or publisher chart is not an independent memory or throughput benchmark.
Run the same acceptance set on a source-linked alternative with aligned prompt budgets and evaluation rules.
Publish both revision IDs, model formats, test harness, pass/fail criteria, observed variance and unsupported cases.
Different engines, sampling strategies, modal input processing and agent permissions confound naive rankings.
Proposed tests, not OMW benchmark results. Read the official model card at the pinned source revision ↗ Inspect OMW field evidence ↗
PRIMARY & INDEPENDENT SOURCES
FAQ
NVIDIA describes a sparse hybrid vision-language model with about 120B total and 12B active parameters, image and video inputs, and text outputs.
Its official model card documents video and image input processing. OMW has not independently run video understanding or temporal-reasoning benchmarks.
The checked config declares 1,048,576 tokens. Serving backends may use shorter configured windows, and OMW has not tested maximum practical input length.
They are separate official weight repositories with different numerical formats and kernel considerations. Equivalent task output is not assumed.
NVIDIA describes native Multi-Token Prediction, which can assist speculative decoding on compatible runtimes. OMW reports no measured speedup.
NVIDIA declares OpenMDW License Agreement v1.1 and describes commercial use subject to its conditions. Read the exact terms before shipping.
MODEL INTELLIGENCE · SOURCE-FIRST
Open Model Weights records Nemotron 3.5 Super VL BF16 as a 120B total / 12B active (publisher-declared)-parameter text, image, video → text open-weight model in the Nemotron 3.5 Super VL family, with NVIDIA listed as the developer. The checked source repository provides 65 recognized weight artifacts, format evidence for Safetensors and a verified context window of 1,048,576 tokens. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.
The repository declares NVIDIA OpenMDW License Agreement v1.1. Commercial and non-commercial use declared by NVIDIA; applicable OpenMDW-1.1 model terms still govern. Open Model Weights reports the checked license evidence and classification; this is not legal advice.
No base model is declared in the checked standard lineage metadata, so Open Model Weights does not infer one. Training recipe status: training-described-in-model-card; data disclosure status: dataset-metadata-declared.
DEPLOYMENT PROFILE
These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.
LICENSE & USE
Commercial and non-commercial use declared by NVIDIA; applicable OpenMDW-1.1 model terms still govern
This section summarizes the checked repository evidence. It is not legal advice.
LINEAGE & PROVENANCE
Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.
RUNTIME SIGNALS
Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.
EVIDENCE TRAIL
This is where the article remains auditable: source links on one side, the registry's observed change history on the other.
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.
01 · WEIGHTS
model-00001-of-00063.safetensorsmodel-00002-of-00063.safetensorsmodel-00003-of-00063.safetensorsmodel-00004-of-00063.safetensorsmodel-00005-of-00063.safetensorsmodel-00006-of-00063.safetensorsmodel-00007-of-00063.safetensorsmodel-00008-of-00063.safetensorsmodel-00009-of-00063.safetensorsmodel-00010-of-00063.safetensorsmodel-00011-of-00063.safetensorsmodel-00012-of-00063.safetensorsmodel-00013-of-00063.safetensorsmodel-00014-of-00063.safetensorsmodel-00015-of-00063.safetensorsmodel-00016-of-00063.safetensorsmodel-00017-of-00063.safetensorsmodel-00018-of-00063.safetensorsmodel-00019-of-00063.safetensorsmodel-00020-of-00063.safetensorsmodel-00021-of-00063.safetensorsmodel-00022-of-00063.safetensorsmodel-00023-of-00063.safetensorsmodel-00024-of-00063.safetensorsmodel-00025-of-00063.safetensorsmodel-00026-of-00063.safetensorsmodel-00027-of-00063.safetensorsmodel-00028-of-00063.safetensorsmodel-00029-of-00063.safetensorsmodel-00030-of-00063.safetensorsmodel-00031-of-00063.safetensorsmodel-00032-of-00063.safetensorsmodel-00033-of-00063.safetensorsmodel-00034-of-00063.safetensorsmodel-00035-of-00063.safetensorsmodel-00036-of-00063.safetensorsmodel-00037-of-00063.safetensorsmodel-00038-of-00063.safetensorsmodel-00039-of-00063.safetensorsmodel-00040-of-00063.safetensorsmodel-00041-of-00063.safetensorsmodel-00042-of-00063.safetensorsmodel-00043-of-00063.safetensorsmodel-00044-of-00063.safetensorsmodel-00045-of-00063.safetensorsmodel-00046-of-00063.safetensorsmodel-00047-of-00063.safetensorsmodel-00048-of-00063.safetensorsmodel-00049-of-00063.safetensorsmodel-00050-of-00063.safetensorsmodel-00051-of-00063.safetensorsmodel-00052-of-00063.safetensorsmodel-00053-of-00063.safetensorsmodel-00054-of-00063.safetensorsmodel-00055-of-00063.safetensorsmodel-00056-of-00063.safetensorsmodel-00057-of-00063.safetensorsmodel-00058-of-00063.safetensorsmodel-00059-of-00063.safetensorsmodel-00060-of-00063.safetensorsmodel-00061-of-00063.safetensorsmodel-00062-of-00063.safetensorsmodel-00063-of-00063.safetensorsmodel-summary-idxs.safetensorsmodel.safetensors.index.json02 · LICENSE
Classification describes the checked license terms; it is not legal advice.
03 · HARDWARE
Weight-only decimal-GB estimate from total parameter count; excludes KV cache, activations, runtime overhead, optimizer state and sharding.
04 · FORMATS
“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.
05 · LINEAGE
06 · TRAINING ASSETS
07 · RUNTIME SUPPORT
Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.
08 · SOURCES
09 · VERIFICATION HISTORY
Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.