NVIDIA Nemotron 3 Super is an open model built for agentic reasoning. Features a 120B hybrid Mamba-Transformer MoE, LatentMoE, and a 1M context window.
Capabilities, design details, and architectural traits
Nemotron 3 Super is an NVIDIA model built for agentic reasoning. Its documented identity centers on a hybrid Mamba-Transformer Mixture-of-Experts design with native long-context processing and built-in speculative decoding.
| Trait | Model-specific fact |
|---|---|
| Hybrid backbone | Uses a hybrid Mamba-Transformer MoE architecture. |
| LatentMoE | Adds LatentMoE for improved accuracy. |
| MTP layers | Includes MTP layers for faster inference through native speculative decoding. |
| Long context | Supports a native context window of up to 1M tokens. |
| Training format | Was pretrained in NVFP4. |
The model is positioned for agentic sessions that need long working context, faster structured generation, and multi-step tool use. Its recipe and deployment docs also show support for inference and customization workflows.
The public release includes pre-trained, post-trained, and quantized checkpoints, plus training data where redistributable. That makes the model more than a single checkpoint drop.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | NVIDIA | Anthropic | Anthropic |
| Release Date | March 11, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.20 Best Input Pricing | $5 | $10 |
| Output Pricing | $0.80 Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | text | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 25.7 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | 37.7 | 78.0 Best Coding Index | 76.5 |
| Agentic Index | 8.8 | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
