Model
Model explorer

Nemotron-4 340B

OPEN
NVIDIA · Nemotron-4 family · released Jun 14, 2024

Open Base/Instruct/Reward family under NVIDIA Open Model License; pioneered large-scale synthetic-data generation (98%+ synthetic alignment).

ReasoningCodingVisionFunction callingTool useAgentic
930.8
Elo · rank #227
Parameters
340B
Active params
340B (dense)
Context
4K tokens
Architecture
Dense Transformer (decoder-only, GQA, RoPE)
License
NVIDIA Open Model License
Languages
50+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning94.3%#11
best: Llama 3.1 405B · 96.9%
BIG-Bench HardReasoning85.4%#17
best: ERNIE 4.5 300B-A47B · 94.3%
HellaSwagReasoning90.5%#6
best: Claude 3 Opus · 95.4%
HumanEvalCoding57.3%#108
best: Claude Opus 4.5 · 99.4%
MMLUKnowledge81.1%#82
best: OpenAI o3 · 92.9%
WinoGrandeReasoning89.5%#3
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
200 GB
VRAM @ FP16
680 GB
Fits on (Q4)
M3 Ultra 512GB
Far exceeds a single RTX 4090's 24GB VRAM even at Q4; NVIDIA sized it to fit one 8-GPU DGX H100 node in FP8.
Quantizations
FP8
Nemotron-4 family
Elo progression across releases
Nemotron-4 15BFeb 2024277.5Nemotron-4 340BJun 2024930.8
Resources
HFnvidia/Nemotron-4-340B-BaseDOCSbuild.nvidia.comAPIAvailable via NVIDIA API
API price weights · each benchmark row carries its own source badge (see methodology)