← Model indexMODEL PROFILE / nemotron-3-5-super-vl

NVIDIA / current

Nemotron 3.5 Super VL

NVIDIA's open vision-language tier spanning enterprise servers down to local RTX devices.

Context128K
ModalityText · Image · Video
AccessWeights · Partners
ReasoningReasoning

Where it fits.

Part of the Nemotron 3.5 family released alongside Bolt edge models, designed for a single model line across cloud and device. Released October 15, 2026 with the Bolt edge family, Super VL targets enterprise servers through to local RTX hardware under NVIDIA's open model licensing.

02

Core capabilities

  • Vision-language reasoning across image and video input
  • Deployment from enterprise servers down to RTX devices
  • Open weights under NVIDIA model licensing
  • Companion Bolt variants for latency-critical edge work
03

Best-fit workflows

  • Vision-language pipelines
  • Edge and RTX deployment
  • Sovereign multimodal systems
04 / OPERATIONS

From evaluation
to production.

Deployment notes

Shortlist for sovereign or edge multimodal pipelines where NVIDIA hardware is already standard. Benchmark real latency and accuracy per device class, and add moderation separately for user-facing surfaces.

Watch before choosing

NVIDIA describes Super VL as matching larger reasoning models at a fraction of the parameters, but publishes limited independent benchmarks. Parameter-efficiency claims are vendor-reported; validate quality on your own multimodal evals before replacing a general frontier model.

05

Family & versions

Specifications and positioning checked against provider documentation.

Open official source↗