Inference
EN
NVIDIA

Nemotron Ultra 253B

Nemotron Ultra 253B is a reasoning model from NVIDIA. 9 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2025-04
RankedEnough results for an index.
IndexAll categories, out of 100.35.9±6.9Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.—Median 67/s
First tokenSeconds to the first output token.—Median 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.32KMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

253.4B parameters
16-bit
533.1 GB
8-bit
283.7 GB
4-bit
162.3 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.1 counted10.4—
CodingCode writing and repair.1 counted44.4—
ReasoningLogic problems and puzzles.1 counted36.8—
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.2 counted38.6204 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted34.049 / 63
MathMath problems.2 counted42.5118 / 168

Results

9 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
MATH-500 Problem SetMath97.0%10 / 27
Instruction-Following EvalInstruction89.5%28 / 70
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for CodeCoding66.3%5 / 12
Graduate-Level Google-Proof Q&AKnowledge76.0%76 / 141
Artificial Analysis GPQA DiamondKnowledge72.8%128 / 199
American Invitational Mathematics Examination 2025Math72.5%35 / 53
Critical Physics TasksReasoning0.0%150 / 214
Artificial Analysis Humanity's Last ExamKnowledge7.4%158 / 222
Artificial Analysis IFBenchInstruction38.2%112 / 146
τ²-Bench Tool-Agent-User EvaluationAgentic11.4%144 / 149

Sources

BenchLMCC BY-NC 4.0

History

5 releases
Index36.4 → 35.9
23 Oct 202612 Nov 2026

Questions

How good is Nemotron Ultra 253B?
Its index is 35.9 out of 100. That is place 234 of 283 ranked models.
What is Nemotron Ultra 253B best at?
Math: 42.5 out of 100, place 118 of 168.
What is the context window of Nemotron Ultra 253B?
32K tokens.
Is Nemotron Ultra 253B open source?
Yes. Its weights are published.
When was Nemotron Ultra 253B released?
April 2025.

Badge

Nemotron Ultra 253B on Inference360: place 234 of 283, index 35.9
HTML
<a href="https://inference360.ai/models/nemotron-ultra-253b"><img src="https://inference360.ai/badges/nemotron-ultra-253b.svg" alt="Nemotron Ultra 253B on Inference360: place 234 of 283, index 35.9" height="20"></a>
Markdown
[![Nemotron Ultra 253B on Inference360: place 234 of 283, index 35.9](https://inference360.ai/badges/nemotron-ultra-253b.svg)](https://inference360.ai/models/nemotron-ultra-253b)

Compare with

Gemini 2.5 Flash35.0GPT-4.1 mini34.5Nemotron 3 Nano 30B32.9

More from NVIDIA

All models
Canary 180M FlashUnrankedAudio Flamingo 3 7BUnrankedParakeet TDT 0.6B v3UnrankedNemotron 3 Nano 30B A3BUnrankedNemotron 3 Nano 30B32.9