NVIDIA Nemotron-3 Super 120B
Open Weights
NVIDIA
NVIDIA's open Nemotron-3 model, a hybrid Mamba-Transformer latent mixture-of-experts with 12B active parameters and a 1M-token context window. Part of NVIDIA's fully open Nemotron family (open weights, training data, and recipes), it targets efficient long-context reasoning.
Strengths
- Fully open weights, training data, and recipes
- Hybrid Mamba-Transformer MoE for efficient long context
- 12B active parameters keep inference efficient
- 1M-token context window
- Optimized for NVIDIA hardware via NIM
Caveats
- Text-only (no vision or audio input)
- Hybrid Mamba-Transformer tooling still maturing
- Best performance tied to NVIDIA serving stack
- Vendor
- NVIDIA
- Context Window
- 1,000,000 tokens
- License
- NVIDIA Open Model License
- Modalities
- text
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
Open weights under the NVIDIA Open Model License - free to self-host; served via NVIDIA NIM and build.nvidia.com.