NVIDIALanguage model
Llama 3.1 Nemotron Ultra 253B
Llama 3.1 Nemotron Ultra 253B has no overall index value yet because there are too few independent benchmark results.
26out of 100+3 vs. last month
based on search, usage and attention
Summary
Llama 3.1 Nemotron Ultra 253B CPT v1 is an NVIDIA base model derived from Llama 3.1 405B Instruct and then continually pretrained. The model card names English and coding languages as its main intended uses. The reasoning checkpoint built on top of it, Llama 3.1 Nemotron Ultra 253B v1, remains a separate model identity and is linked there as a preview API.
The CPT v1 checkpoint has 253 billion parameters, supports up to 131,072 context tokens, and produces text. The open weights use NVIDIA's Open Model License. This base model is intended for further adaptation, not as a finished reasoning chatbot.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 8 published benchmark results
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
MMLU (Show measurement, test conditions, and source)
- Source value
- 88.09
- Score
- 88.09
- Metric
- macro
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA Llama-3.1-Nemotron-Ultra-253B-CPT-v1 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact CPT-v1 card row; macro metric. Card does not publish prompting, evaluation library version, or score units.
MATH500 (Show measurement, test conditions, and source)
- Source value
- 80.4
- Score
- 80.4
- Metric
- micro
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA Llama-3.1-Nemotron-Ultra-253B-CPT-v1 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact CPT-v1 card row; micro metric. The card does not disclose prompt or decoding parameters.
HumanEval (Show measurement, test conditions, and source)
- Source value
- 88.41
- Score
- 88.41
- Metric
- pass@1
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA Llama-3.1-Nemotron-Ultra-253B-CPT-v1 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact CPT-v1 checkpoint; pass@1. Card does not specify code-generation prompt or sampling settings.
GSM-8K (Show measurement, test conditions, and source)
- Source value
- 84.99
- Score
- 84.99
- Metric
- strict-match
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA Llama-3.1-Nemotron-Ultra-253B-CPT-v1 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact CPT-v1 checkpoint; strict-match metric. Card omits shot count and prompting setup.
RULER (Show measurement, test conditions, and source)
- Source value
- 83.21
- Score
- 83.21
- Metric
- unknown
- Unit
- No unit provided
- Benchmark version
- 128K
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA Llama-3.1-Nemotron-Ultra-253B-CPT-v1 model card
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact CPT-v1 checkpoint; 128K RULER setting. Card does not state task mix or scoring unit.
Text Arena Overall (Show measurement, test conditions, and source)
- Source value
- 1348
- Score
- 1348
- Metric
- Leaderboard score
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Arena.ai Text Arena Overall leaderboard
- Status
- active
- Retrieved at
- 2026-10-23
- Methodology
- Official benchmark leaderboard. Text Arena Overall page, snapshot 20 October 2026; the row names Llama 3.1 Nemotron Ultra 253B v1 and lists Nvidia · Nvidia Open Model. The exact serving weight revision, prompt and decoding settings, and model-specific evaluation count are not stated.
- Context
- The page displays ± beside the score but does not define an interval or its coverage. The Votes column is leaderboard votes, not evaluation items, so a model-specific N is not stated. The score unit and a dated protocol version are not defined on this page.
Text Arena Coding (Show measurement, test conditions, and source)
- Source value
- 1391
- Score
- 1391
- Metric
- Leaderboard score
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Arena.ai Text Arena Coding leaderboard
- Status
- active
- Retrieved at
- 2026-10-23
- Methodology
- Official benchmark leaderboard. Text Arena Coding page, snapshot 20 October 2026; the row names Llama 3.1 Nemotron Ultra 253B v1 and lists Nvidia · Nvidia Open Model. The exact serving weight revision, prompt and decoding settings, and model-specific evaluation count are not stated.
- Context
- The page displays ± beside the score but does not define an interval or its coverage. The Votes column is leaderboard votes, not evaluation items, so a model-specific N is not stated. The score unit and a dated protocol version are not defined on this page.
Text Arena Longer Query (Show measurement, test conditions, and source)
- Source value
- 1342
- Score
- 1342
- Metric
- Leaderboard score
- Unit
- No unit provided
- Direction
- Higher is better
- Source type
- official-leaderboard
- Evaluator
- Arena.ai Text Arena Longer Query leaderboard
- Status
- active
- Retrieved at
- 2026-10-23
- Methodology
- Official benchmark leaderboard. Text Arena Longer Query page, snapshot 22 October 2026; the row names Llama 3.1 Nemotron Ultra 253B v1 and lists Nvidia · Nvidia Open Model. The exact serving weight revision, prompt and decoding settings, and model-specific evaluation count are not stated.
- Context
- The page displays ± beside the score but does not define an interval or its coverage. The Votes column is leaderboard votes, not evaluation items, so a model-specific N is not stated. The score unit and a dated protocol version are not defined on this page.
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
Llama 3.1 Nemotron Ultra 253BClaude Mythos 5.1 26Command A 26Grok 4.20 Multi-Agent 26Reka Flash 3 26
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Llama 3.1 Nemotron Ultra 253B, chosen by model line, developer, size, price and release date.
- Nemotron 3 SuperNVIDIA · same developer
- Llama 3.1 Nemotron 70B InstructNVIDIA · same developer
- Llama 3.3 Nemotron Super 49B v1NVIDIA · same developer
- Llama 3.3 Nemotron Super 49B v1.5NVIDIA · same developer
- Nemotron 3 NanoNVIDIA · same developer
- Nemotron Nano 2 9BNVIDIA · same developer
- Mistral NemotronNVIDIA · same developer
- Nemotron Nano 2 VL 12BNVIDIA · same developer
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about Llama 3.1 Nemotron Ultra 253B
How large is the context window of Llama 3.1 Nemotron Ultra 253B?
131,072 tokens. That is how much text Llama 3.1 Nemotron Ultra 253B reads at once, about 200 A4 pages.
Page count as a rule of thumb from Artificial Analysis
Can I run Llama 3.1 Nemotron Ultra 253B locally?
Yes, the weights are open. With 253B parameters it needs a lot of memory, more a server than a laptop.
Stated by NVIDIA
- Open weights
- yes
- License
- NVIDIA Open Model License
- Parameters
- 253B
- active per token
- n/a
Sources and data date
Show all 12 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
| Evaluator | Test series | Composite indices | Recorded active measurements |
|---|---|---|---|
| LMArena | 0 | 0 | 3 |
Without vendor claims and historical or disputed results; those stay labelled in the single values. 3 measurements are not yet assigned to a test series.