NVIDIA Nemotron 3 Super is a 120B-parameter open hybrid MoE model, activating just 12B parameters for maximum compute efficiency and accuracy in complex multi-agent applications. Built on a hybrid Mamba-Transformer...
Specifications
| Provider | NVIDIA |
|---|---|
| Intelligence Index | Not yet benchmarked |
| Context window | 262,144 tokens (262K) |
| Input price | $0.08 / 1M tokens |
| Output price | $0.45 / 1M tokens |
| Input modalities | text |
| Output modalities | text |
| Model ID | nvidia/nemotron-3-super-120b-a12b |
Pricing is per 1M tokens, indicative pricing via OpenRouter (updated 2026-10-09); a provider's own list price may differ. Intelligence Index via Artificial Analysis.
Part of NVIDIA's Nemotron family
NVIDIA's model card for Nemotron 3 Ultra says that Ultra, like the Super model, incorporates Multi-Token Prediction (MTP) layers. The same card lists Ultra at 550B parameters in total with 55B active; compare it with this page's Specifications table if you are choosing between sizes.
Sources
Editorial last reviewed 2026-10-09.
Nemotron 3 Ultra
Frequently asked questions
How much does Nemotron 3 Super cost?
Indicative pricing (via OpenRouter, updated 2026-10-09) is $0.08 per 1M input tokens and $0.45 per 1M output tokens. A provider's own list price may differ. On Vincony, Nemotron 3 Super is available on credit-based pricing alongside 750+ other models.
What is Nemotron 3 Super's context window?
Nemotron 3 Super supports a context window of up to 262,144 tokens (262K), which is how much text it can consider at once.
Who makes Nemotron 3 Super?
Nemotron 3 Super is built by NVIDIA. You can access it — and compare it against other providers' models — through Vincony's multi-model platform.
How do I use Nemotron 3 Super?
You can use Nemotron 3 Super directly through NVIDIA, or via Vincony, which aggregates 750+ models (including Nemotron 3 Super) into one interface with side-by-side comparison and credit-based pricing.