Skip to content

NVIDIA · Chat

Nemotron 3 Ultra

NVIDIA 🇺🇸ChatNemotron familyOpen weights. One of its sources says the weights are publicly available.The maker lists this model as deprecated (7 Nov 2026).
From Requesty's listing:
NVIDIA Nemotron 3 Ultra is NVIDIA's strongest open-weights reasoning model, positioned near GPT-5.4 Mini (xhigh) and ahead of DeepSeek V4-Flash and Qwen3.5-397B-A17B.
llminfo IQ
124.1
Market median price
$1.00 blended / 1M
Max context
1M
Listings
67

Evidence and sources

🤗 nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 Model architecture, as its sources state it llminfo IQ 124.1, ranked among chat models. From 172 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 100.9, 95% interval 89–112. From 13 benchmark families across 6 sources. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.How Coding IQ is computedPoor value (33.1). Costs more per task than its benchmark results predict. From 4 benchmarks with measured task costs. A fair amount of data behind it.How Value for money worksYap score: tokens used per coding task vs. models with the same resultsTypical (53 ±9). Generates about as many tokens as models with similar coding results. Measured with reasoning off. Little data behind it, so treat it as a rough estimate.Median 26,789 output tokens per task with reasoning off.How the Yap score worksReleased 4 Jun 2026Maker HQ: US (source)
All rankings: #7 on OpenRouter today #8 on OpenRouter this month #11 on OpenRouter this week #129 on NanoGPT recentlySources: LMArena data, LiteLLM data, Models.dev data, Morpheus Inference Marketplace data, Requesty data, llminfo researchPrimary data from LMArena from LiteLLM: deprecation date from Models.dev: weights availability from Morpheus Inference Marketplace: architecture from Requesty: description from llminfo research: release date

Advantages and trade-offs

Compared with 10 chat models, picked mainly by popularity and IQ

Advantages

Open weights available

The weights are public. 2 of the 9 peers that say either way have open weights.

Trade-offs

Deprecated by its maker

The maker has already deprecated this model (7 Nov 2026).

We only make a claim when there's enough data on models of the same type. Updated 22 Nov, 06:07 UTC.

Input & output

as stated by the model's providers
Text
Images
Text

Capabilities

Supported operations and provider-stated features

VisionReasoningFunction callingStructured outputPrompt cachingWeb search

Listings

 Aihubmixnemotron-3-ultra-550b-a55b-free · freeSource: AIHubMix data docs1M$0.00$0.00——
 Flowdotnvidia/nemotron-3-ultra-550b-a55b:free · freeSource: FlowDot data1M$0.00$0.00——
 Kilonvidia/nemotron-3-ultra-550b-a55b:free · freeSources: Kilo AI Gateway data, Models.dev data1M$0.00$0.00——
 OpenRouterInference by NVIDIA NIMnvidia/nemotron-3-ultra-550b-a55b:free · freeSource: OpenRouter data docs1M$0.00$0.00541,334.5ms
 Bothubnemotron-3-ultra-550b-a55b:free · freeSource: Models.dev data1M$0.00$0.00——
 Kenarinemotron-3-ultra-550b-a55b:free · freeSource: Models.dev data1M$0.00$0.00——
 Unorouternemotron-3-ultra-550b-a55b:free · freeSource: Models.dev data1M$0.00$0.00——
 Ollamanemotron-3-ultraSource: Models.dev data262K$0.10$3.00——
 Routing Runnemotron-3-ultraSource: Models.dev data131K$0.10$0.10——
 NanoGPTInference by OllamaRegions: UNKnvidia/nemotron-3-ultra-550b-a55bSource: NanoGPTNot published by the provider$0.105$3.15——

5 listings without a price are hidden.

Benchmarks

Named effort levels run from lowest to highest. Select a result to see its source details.

Human preference

Arena ELO44 related benchmarks · 44 results
Source: LMArena data · Nov 2026
BenchmarkNVFP4
Arena ELO
Chinese
Creative Writing
English
Show 40 more benchmarksHide 40 benchmarks
BenchmarkNVFP4
Exclude Ties
Expert
French
German
Hard Prompts
Hard Prompts English
Industry Business and Management and Financial Operations
Industry Entertainment and Sports and Media
Industry Life and Physical and Social Science
Industry Mathematical
Industry Medicine and Healthcare
Industry Software and It Services
Industry Writing and Literature and Language
Korean
Longer Query
Multi Turn
Non English
Polish
Russian
Spanish
Style-Controlled
Text Factuality
Text Factuality - Creative Writing
Text Factuality - English
Text Factuality - Exclude Ties
Text Factuality - Expert
Text Factuality - Hard Prompts
Text Factuality - Hard Prompts English
Text Factuality - Industry Business and Management and Financial Operations
Text Factuality - Industry Entertainment and Sports and Media
Text Factuality - Industry Life and Physical and Social Science
Text Factuality - Industry Medicine and Healthcare
Text Factuality - Industry Software and It Services
Text Factuality - Industry Writing and Literature and Language
Text Factuality - Longer Query
Text Factuality - Multi Turn
Text Factuality - Non English
Text Factuality - Russian
Design Arena EloSource: Design Arena Leaderboards data · Nov 2026

General

Agentic

Arena AgentSource: LMArena data · Nov 2026
BrowseComp2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · official table
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
FinanceAgent v1.1NeMo Evaluator SDK · Vals.ai Financial Agent 1.1 · Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
GDPval2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · official table
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
TAU3-Bench2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · TauBench v3 · four-domain average
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Coding

Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · v6
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SciCode2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · subtask
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
ValsSource: Vals AI

Multilingual

EKI EstonianSource: Keelemudelite mõõdupuu data docs · Aug 2026
WMT24++2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · en→xx
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Reasoning

ApexSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · no tools
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Factuality

Instruction following

IFBench2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · prompt loose
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Long context

Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · official table
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
LongBench v22 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · ≤1M
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
RULERSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
RULER · 1MNeMo Evaluator SDK · 1M · Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.

Math

Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Science

GPQA2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · no tools
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Knowledge

MMLU-Pro2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · official table
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
MMLU-ProX2 results
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · average en/de/fr/es/it/ja/zh/hi/pt/ko
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Other

Apex ShortlistNeMo Evaluator SDK · Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
IOI 2025NeMo Evaluator SDK · official table · Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
PinchBench3 related benchmarks · 3 results
v2 · Best Success Rate
Reported by PinchBench · via BenchBoard docs
OpenClaw · PinchBench v2.0.0 · task count: 147 on live leaderboard / 148 in v2.0.0 release notes · best run
v2 · Average Success Rate
Reported by PinchBench · via BenchBoard docs
OpenClaw · PinchBench v2.0.0 · task count: 147 on live leaderboard / 148 in v2.0.0 release notes · average across 5 runs
Version/Metric Unspecified
Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
NeMo Evaluator SDK · official table · version/metric not stated
τ³-BankingNeMo Evaluator SDK · TauBench v3 · Reported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
τ³-benchReported by NVIDIA · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.

Data credits: NVIDIA, Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning, arXiv:2606.15007v1, Table 10, CC BY 4.0. AA-Omniscience benchmark by Artificial Analysis, Jackson et al. (2025), arXiv:2511.13029. Acquired from the NVIDIA article, not the benchmark leaderboard. (licence) · SEA-HELM leaderboard, AI Singapore (AI Products team), National University of Singapore. Published aggregate scores; results licence not stated (SEA-HELM code is MIT). · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · LMArena, CC BY 4.0 (licence) · Baromètre de l’écriture IA, Johannes Eva; public numeric ranking data, © 2026 Johannes Eva, no results licence stated · BullshitBench by Peter Gostev, MIT (licence) · Clinical Benchmarks (clinicalbenchmarks.ai), CC BY 4.0; fractions converted to percentages; source documents retain their own terms (licence) · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · Keelemudelite mõõdupuu: Independent LLM leaderboard for Estonian. Eesti Keele Instituut. https://moodupuu.eki.ee, CC BY 4.0. Latest per-benchmark results selected with partial-run flags retained. (licence) · HalluHard: Fan, Delsad, Flammarion and Andriushchenko (EPFL), arXiv:2602.01031 · Hugging Face official benchmark leaderboards; licence varies by benchmark · SnakeBench (Greg Kamradt / ARC Prize Foundation), MIT (licence)

Also known as: accounts/fireworks/models/nemotron-3-ultra-nvfp4, azure_ai/fw-nemotron-3-ultra-nvfp4, baseten/nvidia/nvidia-nemotron-3-ultra-550b-a55b, deepinfra/nemotron-3-ultra-550b-a55b, deepinfra/nvidia/nvidia-nemotron-3-ultra-550b-a55b, doubleword/nvidia-nemotron-3-ultra, doubleword/nvidia-nemotron-3-ultra:flex, fireworks/nemotron-3-ultra-nvfp4, fireworks_ai/accounts/fireworks/models/nemotron-3-ultra-nvfp4, fireworks_ai/nemotron-3-ultra-nvfp4, gmi/nvidia/nemotron-3-ultra-550b-a55b, nebius/nvidia/nemotron-3-ultra-550b-a55b, nemotron-3-ultra-550b-a55b, nemotron-3-ultra-550b-a55b-free, nemotron-3-ultra-550b-a55b:free, nemotron-3-ultra-free, nemotron-3-ultra-nvfp4, nvidia nemotron 3 ultra, nvidia-nemotron-3-ultra, nvidia-nemotron-3-ultra-550b-a55b, nvidia-nemotron-3-ultra-550b-a55b-nvfp4, nvidia/nemotron-3-ultra-550b-a55b, nvidia/nemotron-3-ultra-550b-a55b:free, nvidia/nemotron-3-ultra-550b-a55b:thinking, nvidia/nvidia-nemotron-3-ultra-550b-a55b, nvidia/nvidia-nemotron-3-ultra-550b-a55b-bf16, nvidia/nvidia-nemotron-3-ultra-550b-a55b-nvfp4, nvidia_nim/nvidia/nemotron-3-ultra-550b-a55b, perplexity/nemotron-3-ultra-550b-a55b, together_ai/nvidia/nemotron-3-ultra-550b-a55b, wandb/nvidia/nvidia-nemotron-3-ultra-550b-a55b

Like this one. Tap or hover any dotted value, label or icon to see where it comes from and how it is measured. open details when tapped or hovered. How this data works