Nemotron 3.5 Super VL lists 120B total and 12B active parameters. Use total stored weights, precision and runtime allocations when planning memory; active parameters alone are not a hardware requirement.
What Nemotron 3.5 Super VL actually reports
NVIDIA's BF16 model card describes a hybrid Mamba-2, attention and mixture-of-experts design with a C-RADIO vision encoder. It supports text, document and video understanding and context up to one million tokens. The card distinguishes full-precision weights from an optimized NVFP4 release.
What this changes for the reader
An active-parameter figure describes routed work, while deployment must manage the complete model representation. Our suggested budget separates weights, vision inputs, cache and runtime overhead. A small active fraction may make compute efficient without making the unmodified model a small laptop download.
A practical evaluation before adoption
Pick one document with a table and one short permitted video. Ask questions whose answers can be checked against a specific page or timestamp. Record the exact weight precision, input processing and context allocation. Treat tool-use evaluation as another layer: the model's proposed action needs a harness and independently checked results.
- Budget total weights and precision.
- Keep visual preprocessing and cache in the estimate.
- Inspect page- or timestamp-grounded answers.
Limits and availability to keep in view
We have not measured GPU memory, quality or throughput. The model card's maximum context is not evidence of identical performance at that length. Verify runtime support and the release's current terms before exposing a service.
Where to go next
Read our local versus hosted task costs for the adjacent workflow decision. Keep this project's conditions separate from other models' results. If a dependency, artifact or service limit is unclear, leave it unresolved in the comparison rather than filling the gap with an assumption.
Frequently asked questions
Does A12B mean the model has only 12B total parameters?
No. NVIDIA lists 120B total parameters and 12B active.
Are BF16 and NVFP4 the same deployment artifact?
No. The card distinguishes reference BF16 weights from the optimized lower-precision release.
Has FyreLinkz independently tested Nemotron 3.5 Super VL?
No. This article reports the linked primary source and proposes an editorial evaluation; it contains no independent performance benchmark.
Sources & further reading
Primary sources checked Oct 18, 2026. Vendor statements are attributed; editorial advice is our own.
- 1
Help us keep this useful. Check contact availability →
