nvidia
NVIDIA: Nemotron 3 Super
nvidia/nemotron-3-super-120b-a12b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
NVIDIA Nemotron 3 Super is a text-only open model whose card describes a 120B-parameter LatentMoE hybrid (Mamba-2, MoE and attention layers) with 12B active parameters and multi-token prediction. NVIDIA positions it for agentic workflows, tool use, RAG and long-context reasoning in seven languages, with reasoning configurable on or off. The OpenRouter catalog currently exposes a 262k-token context, tool calling and structured outputs, while the card claims up to 1M tokens. Reported benchmark results are vendor-measured; independent validation for specific workloads is advisable.[1][2]
Strengths
- Vendor-described LatentMoE design activates 12B of 120B total parameters, which NVIDIA says targets compute efficiency for high-volume, multi-agent workloads.[1][2]
- Reasoning is configurable: the card toggles it via a chat-template flag, and OpenRouter exposes low, medium and none effort settings, enabled by default.[1][2]
- Long-context focus: the card claims up to 1M tokens and reports evaluations out to 512k tokens, while OpenRouter lists a 262k-token context.[1][2]
Limitations
- Evidence is catalog- and vendor-card-only: benchmarks are NVIDIA-measured and OpenRouter's third-party index coverage is sparse, so workload-specific validation is needed.[1][2]
- Context availability differs by source: OpenRouter lists 262k tokens versus the card's 1M claim, so actual limits depend on the deployment.[1][2]
- The model is text-only, and per the card self-hosting is demanding, with a stated minimum of two H100-80GB GPUs.[1][2]
Where to use it
- Candidate for agentic and tool-calling pipelines, matching NVIDIA's stated optimization for collaborative agents; verify on your own tasks.[1][2]
- Plausible fit for RAG and long-document reasoning given the card's positioning; test retrieval quality in your domain.[2]
- Potential option for high-volume automation such as IT ticket handling across the seven supported languages; pilot before rollout.[2]
Sources
- [1] OpenRouter catalog · NVIDIA: Nemotron 3 SuperSources checked: Nov 13, 2026
- [2] Model documentation · NVIDIA: Nemotron 3 SuperSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
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
- Model ID
- nvidia/nemotron-3-super-120b-a12b
- Canonical ID
- nvidia/nemotron-3-super-120b-a12b-20230311
- Creator
- nvidia
- Added to OpenRouter
- 11.03.2026
- Context length
- 262,144
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.075 / out $0.400
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: optional · default medium · efforts: medium, low, none · parameters: include_reasoning, reasoning, reasoning_effort |
| Coding | CC — below the B band | — |
| Agents | CC — below the B band | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: optional · default medium · efforts: medium, low, none · parameters: include_reasoning, reasoning, reasoning_effort |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Coding index | 37.7 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Agentic index | 1.7 | CC — below the B band | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| DeepInfrabf16 | 262,144 | $0.075 | $0.400 | — | — | 100% |
| DekaLLMfp8 | 262,144 | $0.080 | $0.450 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.