Nemotron 3 Ultra released
NVIDIA · Model release
Source confirmed
Data as of October 14, 2026 · How the score is built
CurrentOpen WeightReasoning
Released Jun 4, 20261M context
Nemotron 3 Ultra
Decision readingNemotron 3 Ultra scores 47.3 out of 100 and ranks #109 of 221. This profile shows 24 source-displayable benchmark rows; its strongest eligible category is Multilingual at #7. Published weights can be self-hosted, but infrastructure cost varies and is not a comparable API token rate.
Released Jun 4, 2026 — see all recent releases
Nemotron 3 Ultra will be repriced, updated or retired. Get each notice with its source and date. Follow model changes
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Capability
47.3/100
field median 50#109 of 221 ranked models
Public
#109of 221
Verified —
Price
Self-hosted; infrastructure cost varies
input median $0.78No comparable first-party hosted token rate
Speed
Not measured
field median 86 tok/sTime to first token not measured
Context
1Mtokens
field median 256,000Reported for this model; direct source link not stored
Strongest published evidence
Multilingual ranks #7. A well-rounded choice across a range of tasks.
Validate before choosing
24 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.
Source-linked · 24 displayable benchmark rows
Follow model changesCategory score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank #95 of 125Percentile 24thWeight 22%6 benchmarksVerified | 28.2 | 6 benchmarks | Verified | |||
| CodingRank #115 of 149Percentile 23rdWeight 20%7 benchmarksVerified | 21.9 | 7 benchmarks | Verified | |||
| ReasoningRank #20 of 29Percentile 32ndWeight 17%2 benchmarksVerified | 70.8 | 2 benchmarks | Verified | |||
| MultimodalWeight 12%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| KnowledgeRank #77 of 181Percentile 58thWeight 12%7 benchmarksVerified | 50.4 | 7 benchmarks | Verified | |||
| MultilingualRank #7 of 17Percentile 63rdWeight 7%1 benchmarkVerified | 89.2 | 1 benchmark | Verified | |||
| Inst. FollowingRank #20 of 144Percentile 87thWeight 5%1 benchmarkVerified | 92.1 | 1 benchmark | Verified | |||
| MathWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured |
24 of 671 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic6/6 verified
- Coding7/7 verified
- Reasoning2/2 verified
- MultimodalNot measured
- Knowledge7/7 verified
- Multilingual1/1 verified
- Inst. Following1/1 verified
- MathNot measured
Verified sourceProvisionalNot measured
Capability shape
Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.
Nemotron 3 Ultra category percentile values
- Agentic24th percentile
- Coding23rd percentile
- Reasoning32nd percentile
- MultimodalNot eligible
- Knowledge58th percentile
- Multilingual63rd percentile
- Instruction following87th percentile
- MathNot eligible
The dashed outline is median of 6 nearest peers.
Eligible category ranks
- Agentic#95/125
- Coding#115/149
- Reasoning#20/29
- MultimodalNot ranked
- Knowledge#77/181
- Multilingual#7/17
- Inst. Following#20/144
- MathNot ranked
Top decileTop quartileMid-fieldNot eligible
Benchmark ledger
Coding opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
Coding7 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| SciCodeScientific Code Benchmark | Score44.6% | Versus best verified row Best verified: Sakana Fugu · 60.1% | Gap15.5 behind | Weight10% ref. weight | Provider exact |
| LiveCodeBench (Vals)LiveCodeBench, Vals AI run | Score86.0% | Versus best verified row Best verified: Claude Fable 5.1 · 90.5% | Gap4.5 behind | Weight8% ref. weight | |
| SWE Multilingual | Score67.7% | Versus best verified row Best verified: Claude Opus 5.5 · 93.9% | Gap26.2 behind | Weight5% ref. weight | Provider exact |
| Terminal-Bench 2.1 | Score56.4% | Versus best verified row Best verified: SWE-2 · 92.8% | Gap36.4 behind | WeightScored in Agentic | Provider exact |
| SWE-bench VerifiedSoftware Engineering Benchmark Verified | Score71.9% | Versus best verified row Best verified: Claude Opus 5 · 96% | Gap24.1 behind | WeightDisplay only | Provider exact |
| LiveCodeBench v6 | Score89.0% | Versus best verified row Best verified: Sakana Fugu-Ultra · 93.2% | Gap4.2 behind | WeightDisplay only | Provider exact |
| SWE-bench (Vals)SWE-bench, Vals AI run | Score69.0% | Versus best verified row Best verified: Claude Opus 5 · 97.0% | Gap28 behind | WeightDisplay only | Verified |
Agentic6 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| Terminal-Bench 2.1 | Score56.4% | Versus best verified row Best verified: SWE-2 · 92.8% | Gap36.4 behind | Weight8% ref. weight | Provider exact |
| BrowseComp | Score44.4% | Versus best verified row Best verified: Atria Dawn Preview · 92.5% | Gap48.1 behind | Weight8% ref. weight | Provider exact |
| Terminal-Bench 2.1 (Vals)Terminal-Bench 2.1, Vals AI run | Score50.9% | Versus best verified row Best verified: GPT-6 Astra · 87.3% | Gap36.4 behind | Weight4% ref. weight | |
| HLE w/ toolsHumanity's Last Exam with tools | Score37.4% | Versus best verified row Best verified: Claude Opus 5.5 · 67.7% | Gap30.3 behind | Weight3% ref. weight | Provider exact |
| τ³-bench resultsτ³-Bench Tool-Agent-User Evaluation | Score70.9% | Versus best verified row Best verified: Mercury 2.5 · 96.0% | Gap25.1 behind | Weight2% ref. weight | Provider exact |
| PinchBench | Score90.0% | Versus best verified row Best verified: Pokee-Isaac 28B · 95.7% | Gap5.7 behind | WeightDisplay only | Provider exact |
Reasoning2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| LongBench v2 | Score61.9% | Versus best verified row Best verified: Qwen3.8 Max · 66.3% | Gap4.4 behind | WeightWeighted 25% | Provider exact |
| CritPtCritical Physics Tasks | Score3.1% | Versus best verified row Best verified: Muse Wonder Full · 32.9% | Gap29.8 behind | WeightDisplay only | Provider exact |
Knowledge7 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| HLEHumanity's Last Exam | Score26.7% | Versus best verified row Best verified: Claude Fable 5.1 · 65% | Gap38.3 behind | Weight44% ref. weight | Provider exact |
| HLE w/o toolsHumanity's Last Exam without tools | Score26.7% | Versus best verified row Best verified: Claude Opus 5.5 · 64.4% | Gap37.7 behind | Weight7% ref. weight | Provider exact |
| MMLU-ProMassive Multitask Language Understanding Professional | Score86.8% | Versus best verified row Best verified: Qwen3.7 Max · 89.6% | Gap2.8 behind | Weight6% ref. weight | Provider exact |
| MMLU-Pro (Vals)MMLU-Pro, Vals AI run | Score85.8% | Versus best verified row Best verified: Claude Fable 5.1 · 92.4% | Gap6.6 behind | Weight6% ref. weight | Verified |
| GPQAGraduate-Level Google-Proof Q&A | Score87% | Versus best verified row Best verified: GPT-6 Astra · 96% | Gap9 behind | Weight3% ref. weight | Provider exact |
| GPQA Diamond (Vals)GPQA Diamond, Vals AI run | Score86.1% | Versus best verified row Best verified: Gemini 3.1 Pro · 95.5% | Gap9.4 behind | Weight2% ref. weight | |
| GPQA-DGPQA Diamond | Score87.0% | Versus best verified row Best verified: GPT-6 Astra · 96.0% | Gap9 behind | WeightDisplay only | Provider exact |
Multilingual1 row
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| MMLU-ProX | Score83% | Versus best verified row Best verified: Qwen3.7 Max · 87% | Gap4 behind | WeightWeighted 100% | Provider exact |
Inst. Following1 row
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| IFBenchInstruction Following Benchmark | Score81.7% | Versus best verified row Best verified: MAI-Thinking-1 · 85% | Gap3.3 behind | WeightWeighted 70% | Provider exact |
Bars run 0–100; the dark tick marks the best source-verified value
All 24 rowsLineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Base entry
Radar
Full release historyNemotron 3 Ultra release history
Radar confirmed these at the source. Use Nemotron 3 Ultra in your work? Explore Radar to follow supported changes and choose your alerts.
Radar
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- Not published
- Context window
- 1M
- Maximum output
- Not sourced yet
- Knowledge cutoff
- Not sourced yet
- Input modalities
- Not sourced yet
- Output modalities
- Not sourced yet
- Parameters
- Not sourced yet
- Availability
- Not sourced yet
- Cloud regions
- Not tracked yet
- Lifecycle
- Current
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing record
- Self-host
- Open weights available; hardware estimate not sourced
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
Nemotron 3 Ultra ranks #109 of 221 on the public leaderboard with a score of 47.25/100. It does not yet have enough sourced coverage for a verified position.
Nemotron 3 Ultra is a open weight model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Official exact-value snapshot from the nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 Hugging Face model card. BenchLM maps directly comparable values from NVIDIA's table and keeps provider-specific rows display-only where appropriate. NVIDIA's earlier announcement used approximate 500B/50B wording; the released checkpoint is 550B total / 55B active.
24 of 671 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Multilingual at #7, while its lowest eligible position is Coding at #115. a well-rounded choice across a range of tasks.
Last updated October 14, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Published weights are available, but BenchLM does not yet have a sourced parameter and VRAM profile for this exact model. Hardware cost estimates stay unavailable until that sizing record is complete.
Estimate VRAM from known parametersQuestions
How does Nemotron 3 Ultra perform overall in AI benchmarks?
Nemotron 3 Ultra ranks #109 out of 221 models on the public BenchAlign leaderboard, with a score of 47.25/100. Its evidence status is Estimated, and this profile shows 24 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.
Is Nemotron 3 Ultra good for knowledge and understanding?
Nemotron 3 Ultra ranks #77 out of 181 eligible models for knowledge and understanding, with a public category score of 50.4/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Nemotron 3 Ultra good for coding and programming?
Nemotron 3 Ultra ranks #115 out of 149 eligible models for coding and programming, with a public category score of 21.9/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Nemotron 3 Ultra good for reasoning and logic?
Nemotron 3 Ultra ranks #20 out of 29 eligible models for reasoning and logic, with a public category score of 70.8/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Nemotron 3 Ultra good for agentic tool use and computer tasks?
Nemotron 3 Ultra ranks #95 out of 125 eligible models for agentic tool use and computer tasks, with a public category score of 28.2/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Nemotron 3 Ultra good for instruction following?
Nemotron 3 Ultra ranks #20 out of 144 eligible models for instruction following, with a public category score of 92.1/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Nemotron 3 Ultra good for multilingual tasks?
Nemotron 3 Ultra ranks #7 out of 17 eligible models for multilingual tasks, with a public category score of 89.2/100. That places it in the current top ten for this category. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Nemotron 3 Ultra open source?
Nemotron 3 Ultra is an open-weight model from NVIDIA. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Does Nemotron 3 Ultra have full benchmark coverage on BenchLM?
No. Nemotron 3 Ultra currently has 51 source-displayable rows across 671 tracked benchmark slots. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Nemotron 3 Ultra?
Nemotron 3 Ultra has a reported context window of 1M in the exact-model catalog record. The value stays visible, but the profile marks its source link as unavailable instead of presenting it as directly documented. Maximum output length remains separate because providers often publish a different limit.