nvidia
NVIDIA: Nemotron 3 Ultra
nvidia/nemotron-3-ultra-550b-a55b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
NVIDIA's Nemotron 3 Ultra (nvidia/nemotron-3-ultra-550b-a55b) is an open-weight, frontier-scale reasoning model with 550B total and 55B active parameters, built on a hybrid Mamba-2, MoE and attention architecture with multi-token prediction. The model card reports toggleable reasoning, tool use and context up to 1M tokens, though the OpenRouter endpoint exposes about 256K. Ten languages are supported, and the OpenMDW-1.1 license permits commercial and non-commercial use. Published benchmarks are vendor-run, so independent validation remains limited.[1][2]
Strengths
- Open weights, training data and recipes released by NVIDIA under the OpenMDW-1.1 license, which the model card states permits commercial and non-commercial use.[2]
- Hybrid LatentMoE design combining Mamba-2, MoE and attention layers with multi-token prediction intended to speed generation; 550B total parameters with 55B active.[1][2]
- Configurable reasoning (on/off plus effort levels per the OpenRouter catalog), with tool calling and structured output support reported by both sources.[1][2]
Limitations
- Demanding serving requirements: NVIDIA lists a minimum of 8 B200/GB200-class or 16 H100 GPUs for this BF16 checkpoint.[2]
- Context figures differ by source: the model card advertises up to 1M tokens while the OpenRouter endpoint exposes 262K, so effective limits may depend on the deployment.[1][2]
- Published benchmark results are vendor-run (NVIDIA's NeMo Evaluator) or catalog-reported; independent verification is limited, so validate on your own workloads.[1][2]
Where to use it
- Complex multi-step agentic and tool-use workflows — NVIDIA's stated design target; treat as a suggested fit pending your own evaluation.[2]
- Long-context analysis and RAG over large documents or codebases, where the deployment's context window allows.[1][2]
- Multilingual reasoning and chat across its ten documented languages, including Spanish and Brazilian Portuguese.[2]
Sources
- [1] OpenRouter catalog · NVIDIA: Nemotron 3 UltraSources checked: Nov 13, 2026
- [2] Model documentation · NVIDIA: Nemotron 3 UltraSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...
Specifications
- Model ID
- nvidia/nemotron-3-ultra-550b-a55b
- Canonical ID
- nvidia/nemotron-3-ultra-550b-a55b-20260604
- Creator
- nvidia
- Added to OpenRouter
- 04.06.2026
- Context length
- 262,144
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.600 / out $2.4
- 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 | CC — below the B band | reasoning: optional · default high · efforts: high, medium, none · parameters: include_reasoning, reasoning, reasoning_effort |
| Coding | BB — next 30% of the cohort | — |
| Agents | BB — next 30% of the cohort | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: optional · default high · efforts: high, medium, 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 | Intelligence index | 11.8 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Coding index | 49.3 | BB — next 30% of the cohort | 13.11.2026 |
| Artificial Analysis | Agentic index | 20.1 | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · 3dwin rate 38% · rank 87 | 1,111Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · asciiartwin rate 36.8% · rank 65 | 1,093Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 35.4% · rank 96 | 1,133Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · datavizwin rate 37.2% · rank 87 | 1,139Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · gamedevwin rate 35.4% · rank 87 | 1,127Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · svgwin rate 32.1% · rank 78 | 1,062Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 35.3% · rank 92 | 1,128Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · websitewin rate 34.4% · rank 102 | 1,131Elo | 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) |
|---|---|---|---|---|---|---|
| DeepInfrafp4 | 262,144 | $0.400 | $1.76 | — | — | — |
| KT | 262,144 | $0.500 | $2.2 | — | — | 100% |
| DeepInfrafp4 | 262,144 | $0.500 | $2.2 | — | — | 100% |
| BaseTenfp4 | 202,800 | $0.600 | $2.4 | — | — | 100% |
| Venicefp8 | 256,000 | $0.625 | $3.13 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.