Model

Nemotron-4 15B

OPEN

NVIDIA · Nemotron-4 family · released Feb 26, 2024

15B multilingual base trained on 8T tokens (50+ languages, 40+ code); beat 4x-larger models on multilingual tasks.

ReasoningCodingVisionFunction callingTool useAgentic

277.5

Elo · rank #351

Parameters

15B

Active params

15B (dense)

Context

4K tokens

Architecture

Dense Transformer (decoder-only, GQA, RoPE, squared-ReLU MLP)

License

NVIDIA Open Model License

Languages

54+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

ARC-ChallengeReasoning55.5%#107

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning80.9%#24

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning58.7%#82

best: ERNIE 4.5 300B-A47B · 94.3%

GSM8KMath46.0%#137

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning82.4%#65

best: Claude 3 Opus · 95.4%

HumanEvalCoding31.6%#151

best: Claude Opus 4.5 · 99.4%

MBPPCoding40.6%#86

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MGSMMath41.3%#48

best: OpenAI o4-mini · 93.7%

MMLUKnowledge64.2%#182

best: OpenAI o3 · 92.9%

PIQAReasoning82.4%#25

best: GPT-4o mini · 93.1%

Social IQaReasoning60.9%#14

best: Apple DCLM-Baseline 7B · 82.9%

WinoGrandeReasoning78.0%#44

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

10 GB

VRAM @ FP16

32 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

—

Nemotron-4 family

Elo progression across releases

Nemotron-4 15BFeb 2024277.5 Nemotron-4 340BJun 2024930.8

Resources

HFnvidia/Nemotron-4-15B-Base DOCSresearch.nvidia.com

API price weights · each benchmark row carries its own source badge (see methodology)