Model

Nemotron Nano 2 (Nemotron-Nano-9B-v2)

OPEN

NVIDIA · Nemotron Nano family · released Aug 18, 2025

Hybrid Mamba-Transformer reasoning model; up to 6x higher throughput than Qwen3-8B in reasoning settings.

ReasoningCodingVisionFunction callingTool useAgentic

1489.1

Elo · rank #122

Parameters

Active params

Undisclosed

Context

128K tokens

Architecture

Hybrid Mamba-2 + Transformer (just 4 self-attention layers), pruned from a 12B/20T-token base

License

NVIDIA Open Model License Agreement

Languages

—

API price (in/out)

$0.04 / $0.16

Modalities

text

Benchmark results

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

AIMEMath72.1%#98

best: GPT-5.2 · 100.0%

BFCL v3Agents66.9%#17

best: Hunyuan-A13B · 78.3%

GPQA DiamondReasoning64.0%#151

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning6.5%#108

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning27.6%#68

best: MiniMax M3 · 83.0%

IFEvalReasoning90.3%#14

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding71.1%#66

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math97.8%#11

best: GPT-5 · 99.4%

Run it locally

VRAM @ Q4

8 GB

VRAM @ FP16

18 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

Q4_K_M GGUF is only ~6.5GB — comfortably fits a single RTX 4090; no specific measured llama.cpp 4090 tokens/sec figure found.

Quantizations

GGUF

Nemotron Nano family

Elo progression across releases

Nemotron Nano 2 (Nemotron-Nano-9B-v2)Aug 20251489.1

Resources

HFnvidia/NVIDIA-Nemotron-Nano-9B-v2 DOCSbuild.nvidia.com/nvidia/nvidia-nemotron-nano-9b-v2/modelcard APIAvailable via NVIDIA API

API price $0.04/$0.16 · each benchmark row carries its own source badge (see methodology)