Research note
Research note: Nvidia releases Nemotron 3.5 Super VL, an open 120-billion-parameter vision model, with a new Indic-language dataset
How we checked Nemotron 3.5 Super VL’s size, openness and Indic data, and separated Nvidia’s benchmark claims.
What we set out to check
The model’s size, licence and availability, the Indic data, and which performance figures come from Nvidia.
What we found
- Nvidia’s blog gives 120B total and 12B active parameters, the modalities and the Indic dataset.
- The blog does not name the licence; VentureBeat reports OpenMDW.
- VentureBeat reads the post-training comparisons from Nvidia’s charts and notes Nvidia has not published test details.
Numbers that differ
Nvidia’s page gives Scale AI’s latency gain as 5x in one place and 10x in an AI-generated summary. We do not use either figure.
Labels we chose and why
- Confirmed for size, openness and the Indic dataset.
- Company claim for benchmarks.
- Unconfirmed for the Bolt E4B date, seen only in VentureBeat.
What we could not verify
- The licence text.
- The context length.
What we searched
- Nvidia Nemotron 3.5 Super VL
- Nemotron 3.5 Super VL licence OpenMDW
- Nemotron Indic dataset 22 languages
Every source we read
| Source | Used? | Why |
|---|---|---|
| Nvidia Technical Blog | USED | Size, modalities, Indic data, benchmark claims. |
| VentureBeat | USED | Licence, pricing for production use, post-training comparisons, Bolt models. |
| Hugging Face | USED | Confirms the published checkpoint name and size. |