NVIDIA model product
Nemotron 3 Ultra is an open NVIDIA large language model with 550B total and 55B active parameters, a hybrid LatentMoE architecture, configurable reasoning, and support for up to a 1M-token context.
Updated Sep 8, 2026. Default version: Nemotron 3 Ultra (550B A55B)
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
Benchmark scores for Nemotron 3 Ultra (550B A55B).
Benchmark | Score | Rank | Participants | Percentile | Evidence | Evaluated |
|---|
| BenchmarkApex | Score84.80% | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkIMO-AnswerBench | Score92.30% | Rank01 | Participants20 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkOmniScience | Score78.70% | Rank01 | Participants3 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkPinchBench | Score90.00% | Rank01 | Participants6 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkProfBench | Score56.00% | Rank01 | Participants4 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkRULER | Score94.70% | Rank01 | Participants4 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkGDPval | Score46.70% | Rank03 | Participants3 | Percentile0.00% | EvidenceC | Evaluated |
| BenchmarkIFBench | Score81.70% | Rank03 | Participants39 | Percentile94.74% | EvidenceC | Evaluated |
| BenchmarkLongBench v2 | Score61.90% | Rank04 | Participants17 | Percentile81.25% | EvidenceC | Evaluated |
| BenchmarkMMLU-ProX | Score83.00% | Rank05 | Participants32 | Percentile87.10% | EvidenceC | Evaluated |
| BenchmarkCritPT | Score3.10% | Rank06 | Participants6 | Percentile0.00% | EvidenceC | Evaluated |
| BenchmarkMulti-Challenge | Score63.80% | Rank06 | Participants29 | Percentile82.14% | EvidenceC | Evaluated |
| BenchmarkWMT24++ | Score83.70% | Rank06 | Participants23 | Percentile77.27% | EvidenceC | Evaluated |
| BenchmarkFinance Agent | Score53.70% | Rank08 | Participants8 | Percentile0.00% | EvidenceC | Evaluated |
| BenchmarkLiveCodeBench v6 | Score89.00% | Rank08 | Participants62 | Percentile88.52% | EvidenceC | Evaluated |
| BenchmarkTAU3-Bench | Score22.60% | Rank08 | Participants8 | Percentile0.00% | EvidenceC | Evaluated |
| BenchmarkAA-LCR | Score65.40% | Rank10 | Participants19 | Percentile50.00% | EvidenceC | Evaluated |
| BenchmarkMMLU-Pro | Score86.80% | Rank10 | Participants138 | Percentile93.43% | EvidenceC | Evaluated |
| BenchmarkSciCode | Score44.60% | Rank10 | Participants24 | Percentile60.87% | EvidenceC | Evaluated |
| BenchmarkFinance Agent v2 | Score37.53% | Rank20 | Participants26 | Percentile24.00% | EvidenceB | Evaluated |
| BenchmarkSWE-bench Multilingual | Score67.70% | Rank26 | Participants43 | Percentile40.48% | EvidenceC | Evaluated |
| BenchmarkTerminal-Bench 2.1 | Score56.40% | Rank30 | Participants35 | Percentile14.71% | EvidenceC | Evaluated |
| BenchmarkLM Arena Agent Tool Hallucination | Score0.23% | Rank33 | Participants49 | Percentile33.33% | EvidenceA | Evaluated |
| BenchmarkHumanity's Last Exam | Score37.40% | Rank42 | Participants103 | Percentile59.80% | EvidenceC | Evaluated |
| BenchmarkLM Arena Text | Score1,445.01 rating | Rank43 | Participants210 | Percentile79.90% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Praise Complaint | Score-14.99% | Rank46 | Participants49 | Percentile6.25% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Steerability | Score-9.71% | Rank46 | Participants49 | Percentile6.25% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Task Outcome Explicit | Score-15.71% | Rank47 | Participants49 | Percentile4.17% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Leaderboard | Score-14.43% | Rank48 | Participants49 | Percentile2.08% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Bash Recovery Steps | Score-31.95% | Rank48 | Participants49 | Percentile2.08% | EvidenceA | Evaluated |
Preference and agent-evaluation results for the default version.
Arena | Category | Rank | Rating / score | Votes | Observations | Result date |
|---|
| Arenatext | Categoryfrench | Rank22 | Rating / score1,476.87 | Votes450 | ObservationsN/A | Result date |
| Arenatext | Categorygerman | Rank22 | Rating / score1,464.66 | Votes202 | ObservationsN/A | Result date |
| Arenatext | Categoryspanish | Rank28 | Rating / score1,454.73 | Votes434 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry life and physical and social science | Rank31 | Rating / score1,469.19 | Votes1,745 | ObservationsN/A | Result date |
| Arenaagent tool hallucination | Categoryoverall | Rank33 | Rating / score0.00 | VotesN/A | Observations150.9K | Result date |
| Arenatext | Categoryexpert | Rank36 | Rating / score1,472.11 | Votes1,143 | ObservationsN/A | Result date |
| Arenatext | Categoryenglish | Rank38 | Rating / score1,454.26 | Votes4,957 | ObservationsN/A | Result date |
| Arenatext | Categorymath | Rank38 | Rating / score1,457.80 | Votes548 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry legal and government | Rank39 | Rating / score1,458.16 | Votes906 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry mathematical | Rank39 | Rating / score1,460.30 | Votes653 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry medicine and healthcare | Rank39 | Rating / score1,476.85 | Votes766 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry software and it services | Rank40 | Rating / score1,469.80 | Votes4,105 | ObservationsN/A | Result date |
| Arenatext | Categorychinese | Rank41 | Rating / score1,491.38 | Votes629 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry medicine and healthcare | Rank42 | Rating / score1,456.64 | Votes800 | ObservationsN/A | Result date |
| Arenatext | Categoryoverall | Rank43 | Rating / score1,445.01 | Votes10,700 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryexpert | Rank43 | Rating / score1,476.18 | Votes1,107 | ObservationsN/A | Result date |
| Arenatext | Categorycoding | Rank45 | Rating / score1,469.01 | Votes2,953 | ObservationsN/A | Result date |
| Arenatext | Categoryexclude ties | Rank45 | Rating / score1,439.83 | Votes8,095 | ObservationsN/A | Result date |
| Arenatext | Categoryhard prompts | Rank45 | Rating / score1,452.19 | Votes7,020 | ObservationsN/A | Result date |
| Arenaagent praise complaint | Categoryoverall | Rank46 | Rating / score-0.15 | VotesN/A | Observations971 | Result date |
| Arenaagent steerability | Categoryoverall | Rank46 | Rating / score-0.10 | VotesN/A | Observations4.2K | Result date |
| Arenaagent task outcome explicit | Categoryoverall | Rank47 | Rating / score-0.16 | VotesN/A | Observations3.8K | Result date |
| Arenatext factuality | Categoryindustry life and physical and social science | Rank47 | Rating / score1,469.60 | Votes1,742 | ObservationsN/A | Result date |
| Arenaagent | Categoryoverall | Rank48 | Rating / score-0.14 | VotesN/A | Observations168.5K | Result date |
| Arenaagent bash recovery steps | Categoryoverall | Rank48 | Rating / score-0.32 | VotesN/A | Observations8.7K | Result date |
| Arenatext | Categoryindustry business and management and financial operations | Rank48 | Rating / score1,432.85 | Votes2,143 | ObservationsN/A | Result date |
| Arenatext | Categorynon english | Rank50 | Rating / score1,427.10 | Votes5,742 | ObservationsN/A | Result date |
| Arenatext | Categoryhard prompts english | Rank52 | Rating / score1,453.82 | Votes3,417 | ObservationsN/A | Result date |
| Arenatext | Categorylonger query | Rank53 | Rating / score1,439.03 | Votes4,845 | ObservationsN/A | Result date |
| Arenatext style control | Categorymath | Rank54 | Rating / score1,444.53 | Votes548 | ObservationsN/A | Result date |
Official vendor API pricing appears first, followed by individual provider offers.
Provider | Provider model ID | Region | Input / 1M | Output / 1M | Context | Updated |
|---|
| ProviderKenari | Provider model IDnemotron-3-ultra-550b-a55b | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderUnoRouter | Provider model IDnemotron-3-ultra-550b-a55b:free | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderRequesty | Provider model IDnemotron-3-ultra-550b-a55b | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| Providerrouting.run | Provider model IDnemotron-3-ultra | Regionglobal | Input / 1M$0.10 | Output / 1M$0.10 | Context131.1K | Updated |
| ProviderNvidia | Provider model IDnvidia/nemotron-3-ultra-550b-a55b | Regionglobal | Input / 1M$0.50 | Output / 1M$2.5 | Context1M | Updated |
| ProviderNanoGPT | Provider model IDnvidia/nemotron-3-ultra-550b-a55b | Regionglobal | Input / 1M$0.50 | Output / 1M$2.5 | Context1M | Updated |
| ProviderDevPass (LLM Gateway) | Provider model IDnemotron-3-ultra-550b | Regionglobal | Input / 1M$0.50 | Output / 1M$2.2 | Context1M | Updated |
| ProviderKilo Gateway | Provider model IDnvidia/nemotron-3-ultra-550b-a55b | Regionglobal | Input / 1M$0.50 | Output / 1M$2.2 | Context256K | Updated |
| ProviderPioneer | Provider model IDnvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 | Regionglobal | Input / 1M$0.50 | Output / 1M$2.5 | Context1M | Updated |
| ProviderTogether AI | Provider model IDnvidia/nemotron-3-ultra-550b-a55b | Regionglobal | Input / 1M$0.60 | Output / 1M$3.6 | Context512.3K | Updated |
| ProviderVercel AI Gateway | Provider model IDnvidia/nemotron-3-ultra-550b-a55b | Regionglobal | Input / 1M$0.60 | Output / 1M$2.4 | Context1M | Updated |
| ProviderFireworks AI | Provider model IDaccounts/fireworks/models/nemotron-3-ultra-nvfp4 | Regionglobal | Input / 1M$0.60 | Output / 1M$2.4 | Context262.1K | Updated |
| ProviderOpenRouter | Provider model IDnvidia/nemotron-3-ultra-550b-a55b | Regionglobal | Input / 1M$0.625 | Output / 1M$3.13 | Context262.1K | Updated |
| ProviderNebius Token Factory | Provider model IDnvidia/Nemotron-3-Ultra-550b-a55b | Regionglobal | Input / 1M$1 | Output / 1M$3 | Context1M | Updated |
Provider-specific output speed and catalog latency for Nemotron 3 Ultra (550B A55B). Runtime does not affect the capability score.
No provider-specific speed or latency record is linked to the default version yet.
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
Technical details for the model's default version.
Available versions of this model. The score column identifies the version used in the overall ranking.
Version | Released | LLMBoard | Parameters | Context | Max output | Open weights | License |
|---|
| VersionNemotron 3 Ultra (550B A55B) | Released | LLMBoard61.30 | Parameters550B | ContextN/A | Max outputN/A | Open weightsYes | LicenseOpenMDW License v1.1 |
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
Key information about Nemotron 3 Ultra and its available data.
1 license. It uses Mamba-2, MoE, and selected Attention layers with Multi-Token Prediction, and supports tool use, long-context analysis, RAG, and 10 languages.
Data as of 2026-09-08.
Common questions about Nemotron 3 Ultra.
Nemotron 3 Ultra's default version was released on Jun 4, 2026.
Nemotron 3 Ultra's official API price is $0.50 per million input tokens and $2.5 per million output tokens via Nvidia. The lowest tracked third-party offer starts at $0.10 input and $0.10 output via routing.run.
Nemotron 3 Ultra was created by NVIDIA.
A context window is not available for the default version.
Yes. The default version is marked as open weight under OpenMDW License v1.1.
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
14 provider offerings are linked to the default version.