NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

Home/Models/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

kind: nvfp4
quant: nvfp4 (4-bit)
family: nemotron
modality: text-generation
root model: nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
provider: nvidia
license: other
language: en, es, fr, de, ja, it
context: 256k
library: transformers
vRAM (GB): 812.7
size on disk: —
downloads: 1,370,051
likes: 171
created: —
last modified: 2026-03-15T04:27:10+00:00 (2026)
huggingface: https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4
tags: transformerssafetensorsnemotron_htext-generationnvidiapytorchconversationalcustom_codeenesfrdejaitdataset:nvidia/Nemotron-Pretraining-Code-v1dataset:nvidia/Nemotron-CC-v2dataset:nvidia/Nemotron-Pretraining-SFT-v1dataset:nvidia/Nemotron-CC-Math-v1dataset:nvidia/Nemotron-Pretraining-Code-v2dataset:nvidia/Nemotron-Pretraining-Specialized-v1dataset:nvidia/Nemotron-CC-v2.1dataset:nvidia/Nemotron-CC-Code-v1dataset:nvidia/Nemotron-Pretraining-Dataset-sampledataset:nvidia/Nemotron-Competitive-Programming-v1

benchmark	category	score	context	conf	evidence
AIME 2025	math	89.1	AIME25 (no tools)	medium	`\| AIME25 (no tools) \| 89.1 \| 87.7 \| 86.7 \|`
MMLU-Pro	knowledge	78.3	MMLU-Pro	medium	`\| MMLU-Pro \| 78.3 \| 78.1 \| 77.4 \|`
GPQA	reasoning	73.0	GPQA (no tools)	medium	`\| GPQA (no tools) \| 73.0 \| 72.5 \| 71.9 \|`
IFBench	instruction-following	71.5	IFBench (prompt)	medium	`\| IFBench (prompt) \| 71.5 \| 72.2 \| 70.7 \|`
LiveCodeBench	coding	68.3	LiveCodeBench (v6 2025-08–2025-05)	medium	`\| LiveCodeBench (v6 2025-08–2025-05) \| 68.3 \| 67.6 \| 65.4 \|`
MMLU-Pro	knowledge	59.5	MMLU-ProX (avg over langs)	medium	`\| MMLU-ProX (avg over langs) \| 59.50 \| 59.6 \| 57.8 \|`
tau-bench	agentic	56.9	TauBench V2 (Retail)	medium	`\| TauBench V2 (Retail) \| 56.9 \| 55.6 \| 54.1 \|`
tau-bench	agentic	49.0	TauBench V2 (Average)	medium	`\| TauBench V2 (Average) \| 49.0 \| 47.0 \| 45.6 \|`
tau-bench	agentic	48.0	TauBench V2 (Airline)	medium	`\| TauBench V2 (Airline) \| 48.0 \| 44.8 \| 41.5 \|`
tau-bench	agentic	42.2	TauBench V2 (Telecom)	medium	`\| TauBench V2 (Telecom) \| 42.2 \| 40.8 \| 41.2 \|`
SciCode	coding	33.0	SciCode (subtask)	medium	`\| SciCode (subtask) \| 33.0 \| 31.9 \| 30.7 \|`
Humanity's Last Exam	knowledge	10.2	HLE (no tools)	medium	`\| HLE (no tools) \| 10.2 \| 10.3 \| 9.4 \|`

none scraped.