Language model
Nemotron 3 Super 120B A12B
NVIDIA
- Released
- March 11, 2026
- Data date
- September 26, 2026
Nemotron 3 Super 120B A12B is NVIDIA’s model with LatentMoE and a hybrid Mamba-2, MoE, and attention architecture. It has 120 billion total parameters and activates 12 billion per token.
NVIDIA advertises up to 1M context tokens. Its standard vLLM example sets the maximum model length to 262,144 tokens. The model card describes an additional long-context configuration for the larger window.
Specifications and access
| Specification | Value and source |
|---|---|
| Weights | Open weightsSource |
| Access | APISource |
| Parameters | 120 billionSource |
| Active parameters | 12 billionSource |
| License | NVIDIA Open Model LicenseSource |
| Inputs | TextSource |
| Outputs | TextSource |
| Reasoning | YesSource |
| Tool calling | YesSource |
| Advertised context | Up to 1M tokensSource |
| Packaged default configuration | 262,144 tokensSource |
Popularity
18 / 100
Sources and data date
| Type | Evidence and data date |
|---|---|
| Additional source | models.dev catalog (retrieved September 26, 2026) |
| Additional source | NVIDIA Nemotron 3 Super 120B A12B model card (retrieved October 2, 2026) · NVIDIA Nemotron 3 Super BF16 model card · Editorial description reviewed October 2, 2026 |
| Additional source | NVIDIA Nemotron 3 Super 120B A12B packaged config (retrieved October 2, 2026) |
| Additional source | Nemotron 3 Super technical report · Editorial description reviewed October 2, 2026 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Creative Commons Attribution 4.0 |