Nemotron Nano 9B v2

Name: Nemotron Nano 9B v2
Author: NVIDIA

NVIDIA · Nemotron 2 · released 2025-08-18 · data checked 2026-11-01 Needs verification

Nemotron Nano 9B v2 is a 8.9B-parameter dense model from NVIDIA (United States), released 2025-08-18 under the NVIDIA Open Model license. It scores 52.5 on the Small LM Index, #11 of 26 small models. It needs about 6.2 GB of memory at 4-bit with an 8K-token context.

Hybrid Mamba-2/Transformer.

Small LM Index

52.5

#11 of 26 scored small models

Likely 47.7–57.1 · from 3 of 6 core benchmarks

Parameters

8.9B

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

128K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 3 reported by NVIDIA. How we label sources.

GPQA Diamond Lab-reported Knowledge64.0

Class best: Qwen3.5 9B 81.7

AIME Lab-reported Math · AIME 202572.1

Class best: Apriel 1.6 15B Thinker 88.0

LiveCodeBench Lab-reported Coding71.1

Class best: Granite 4.2 8B 73.2

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

MMLU-Pro≈ 74

Humanity's Last Exam≈ 7

SWE-bench Verified≈ 34

Specifications

Lab: NVIDIA (US · United States)
Architecture: Dense transformer
Total parameters: 8.9B
Active parameters: All (dense)
Size class: Small
Context window: 128K tokens
Input modalities: text
Reasoning: Hybrid (thinking on/off)
License: NVIDIA Open Model
Openness: Open weights
Memory to run: ~6.2 GB at 4-bit, 8K context (more sizes)
Hugging Face: 141.1K downloads (30 days) · 532 likes

Weights on Hugging Face ↗

Compare with

Nemotron Nano 9B v2 vs Qwen3.5 9B (9B, index 63.9)
Nemotron Nano 9B v2 vs MiniCPM-SALA (9.5B, index 52.6)
Nemotron Nano 9B v2 vs Apertus 1.5 8B (8.9B, index 20.1)
Nemotron Nano 9B v2 vs DeepSeek-R1-0528-Qwen3-8B (8.2B, index 53.1)

Sources

https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2

Something wrong here? Report it.

Run it locally

Uses bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ2_M, IQ2_S, IQ3_M, IQ3_XS, IQ3_XXS, IQ4_NL, IQ4_XS….

Ollama

ollama run hf.co/bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF:Q4_K_M

llama.cpp

llama-server -hf bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

vLLM (GPU server)

vllm serve nvidia/NVIDIA-Nemotron-Nano-9B-v2

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	128K context
16-bit F16	19 GB	19 GB	21 GB
8-bit Q8_0	10 GB	11 GB	12 GB
4-bit Q4_K_M	6.1 GB	6.6 GB	8.2 GB

Check it against your GPU, Mac or phone →