whatmodel.sh / Nemotron Nano 9B v2

Nemotron Nano 9B v2

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Nemotron Nano 9B v2 is an open-weights language model from NVIDIA, released August 2025. Its published size is 8.89B parameters and a 131k-token context window. Licence: nvidia-open-model.

55.7±0.7#254 of 719 LLMsfrom 34 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Nemotron Nano 9B v2, out of 828 with a score, then more from NVIDIA.

#modelscorewhereRAMreleased
303DeepSeek R1 Distill Qwen 14B DeepSeek56±1local~10GB2025-01
304Ministral 3 14B Mistral56±1api $0.20/M +local~9.1GB2025-12
305Nemotron Nano 9B v2 NVIDIA56±1api $0.10/M +local~7.6GB2025-08
306Magistral Medium 1.0 (2506) Mistral56±1comparison only2025-06
307GPT-4 Turbo (2024-04-09) OpenAI55±0api $15.00/M2024-04
more from NVIDIA
98Nemotron 3 Ultra 550B-A55B NVIDIA69±1api $1.35/M +local~89GB2026-06
155Nemotron Cascade 2 30B-A3B NVIDIA66±1local~20GB2026-03
168Nemotron 3 Super 120B-A12B NVIDIA65±1api $0.28/M +local~73GB2026-03

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #178 of 300 * · 1.1% of its score · 1 test

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Mellum2 12B-A2.5B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
BFCL v3 lab +266.9% ✎▁▁▁▁▁▃▅█▆▅▄▁ beats 132/170medium · avg 58%▇ 1.1%blunt · at its level · small run · several runs
coding▁▁▁▃▆██▇▆▄▃▁ #217 of 458 · 23% of its score · 7 tests

Beats Qwen3.5 4B, Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B) here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 3B Reasoning 2512, DeepSeek-Coder-V2 Instruct (0724) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LiveCodeBench Pro 2025Q2 (Medium) lab4.8% ✎▃▃····▃·█·▃▃ beats 2/7very hard · avg 2.4%▇▇▇▇ 8.8%sharp · too hard for it
LiveCodeBench Pro 2025Q2 (Easy) lab59.3% ✎▄·▄····▄▄▄█▄ beats 2/7medium · avg 45%▇▇ 5.9%at its level
LiveCodeBench v6 (Aug 2024–May 2025) lab65.3% ✎▄▄···▄·▄▄█·▄ beats 2/7medium · avg 57%▇ 2.6%at its level
HumanEval+89.7% ✎▂·······▂·▂█ beats 3/6easy · avg 80%▇ 1.7%sharp
MBPP+66.1% ✎▄▄····▄·▄··█ beats 3/5medium · avg 62%▇ 1.6%at its level
LiveCodeBench v383.4% ✎▃▃·····▃·▃·█ beats 5/6easy · avg 65%▇ 1.3%typical
LiveCodeBench (version unspecified) lab +171.1% ✎▁···▁▁▃▅▅█▄▅ beats 31/37medium · avg 44%▇ 1.2%at its level · small run
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #181 of 393 · 3.5% of its score · 4 tests

Beats Seed-OSS 36B Instruct, Nemotron-Cascade-14B-Thinking, Granite 4.2 8B here, though they rank higher overall. Trails LFM2.5 1.2B Instruct, LFM2.5 1.2B Thinking, Gemma 3 12B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AlpacaEval 2 length-controlled58.0% ✎█···█·█·█·██ beats 3/5medium · avg 44%▇ 2.4%at its level
IFEval prompt-level strict lab86.1% ✎▁▁··▂··▂▇█▃▃ beats 16/26easy · avg 77%▇ 0.9%small run
IFEval (aggregation unspecified)86.0% ✎▄▄·····█▄▄·▄ beats 3/6easy · avg 79%▇ 0.3%noisy
IFEval instruction-level strict lab90.3% ✎▄▄····▄▄█▄·█ beats 8/8—not counted yet: no other model in the same run to compare it with
math▁▂▃▅▆▅▄█▄▁▁▁ #163 of 398 · 33% of its score · 5 tests

Beats MiniMax-M2.7, MiniMax-M2.1, Mistral Medium 3.5 (128B) here, though they rank higher overall. Trails Olmo 3 7B Think here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AIME 2025 +358.9% ✎▄··▄····█▄▄▄ beats 3/6medium · avg 51%▇▇▇▇▇▇ 15%sharp · at its level · several runs
AIME 2024 +172.1% ✎▄·····▄··▄██ beats 4/6medium · avg 59%▇▇▇▇ 10%sharp · at its level · several runs
OMEGA (math generalisation)42.4% ✎▄······█▄▄·█ beats 5/6hard · avg 30%▇ 3.3%at its level
MATH (subset unspecified)94.4% ✎▃·····▃···▅█ beats 2/6very easy · avg 90%▇ 3.0%maxed out
MATH-500 lab +197.8% ✎▁▃▂▃▅▇▅▇█▅▂▁ beats 73/84very easy · avg 92%▇ 1.6%maxed out · small run
other tests12 not in a task · 39% of its score
testresultvs the others in its rundifficultycountswhy
GPQA (subset unspecified) +256.2% ✎▃······▃·▅·█ beats 4/6hard · avg 43%▇▇▇▇▇ 12%sharp · at its level · several runs
MMLU +184.3% ✎▄···▄··█··▄█ beats 4/6easy · avg 77%▇▇ 5.9%several runs
BIG-Bench Hard86.2% ✎▄·····▄·▄·██ beats 3/6easy · avg 78%▇▇ 5.7%sharp
GPQA Diamond lab64.0% ✎▁▁▁▁▂▄▅▅███▄ beats 124/310hard · avg 46%▇▇ 4.6%sharp · at its level · small run
IFBench (aggregation unspecified) +134.6% ✎█·██·██····█ beats 2/5hard · avg 31%▇ 2.5%at its level · small run · several runs
AGIEval English83.1% ✎▄····▄·▄▄··█ beats 3/5easy · avg 76%▇ 2.3%typical
MMLU-Pro lab73.3% ✎▁▁·▁▁▁▃▂▄▅█▁ beats 95/188medium · avg 60%▇ 2.3%at its level · small run
ZebraLogic60.8% ✎█▄····▄·▄·▄▄ beats 3/6medium · avg 53%▇ 1.6%at its level
4 more tests
testresultvs the others in its rundifficultycountswhy
PopQA17.9% ✎▄··█···▄···█ beats 1/5hard · avg 22%▇ 1.3%blunt · at its level
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy62.4% ✎▃·▁▁▂▁▂▅█▇▄▁ beats 37/100medium · avg 68%▇ 0.6%at its level · noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy92.1% ✎▁▁▁▁▁▁▁▂▄█▄▃ beats 125/146easy · avg 83%▇ 0.3%noisy
Humanity's Last Exam lab6.5% ✎▁▁▁▃▆▆▇██▆▁▃ beats 7/104very hard · avg 3.8%—not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · Amazon Bedrock · $0.10/M · $0.06 in / $0.23 out per 1M tokens · 131k context · docs.aws.amazon.com ↗ · price sourceUSD per 1M tokens, Amazon Bedrock on-demand 'NVIDIA Nemotron Nano 2' (Bedrock model ID nvidia.nemotron-nano-9b-v2; the separate 'Nano 2 VL' row is the 12B VL model), US East (N. Virginia)/Ohio and US West (Oregon). GovCloud rates excluded. Currency: USD. Amazon Bedrock official pricing page, observed 2026-09-25.

llama.cpp · recommended

version

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Nemotron Nano 9B v2 · needs ~5.83 GB · from huggingface.co/bartowski
    chat in the terminal
    llama cli -hf bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Nemotron Nano 9B v2 · needs ~5.83 GB · from huggingface.co/bartowski

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Nemotron Nano 9B v2 · needs ~5.83 GB · from huggingface.co/bartowski
    chat in the terminal
    ollama run hf.co/bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF:Q4_K_M

MLX (Mac)

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Nemotron Nano 9B v2 · needs ~7.2 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/NVIDIA-Nemotron-Nano-9B-v2-6bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/NVIDIA-Nemotron-Nano-9B-v2-6bit

Official weights and documentation: huggingface.co/nvidia ↗

$ how the score was worked out · Olmo 3 41%, NVIDIA — Nemotron-Cascade: Sc… 39%, NVIDIA vendor-reported benchm… 13% and 2 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Olmo 314▇▇▇▇▇▇▇▇▇▇▇▇ 41%CC BY 4.0
NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models · lab-reported11▇▇▇▇▇▇▇▇▇▇▇▇ 39%CC BY 4.0
NVIDIA vendor-reported benchmark scores for its own Nemotron models (Hugging Face model cards) · lab-reported5▇▇▇▇ 13%Factual vendor-reported score (not licensed content)
allenai/Olmo-3-7B-Think model card2▇▇ 6%Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard2▇ 1%MIT
prior · pull toward the middle▇ 0%
weighted average= 55.7

not counted (7)

nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardAIME 2025 [correct]72.1%the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardMATH-500 [correct]97.8%the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardGPQA (subset unspecified) [correct]64.0%the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardLiveCodeBench (version unspecified) [correct]71.1%the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardBFCL v3 [correct]66.9%the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts
nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardIFEval instruction-level strict [correct]90.3%fewer than three models have this test yet, or this source ran it for this model alone
nvidia/NVIDIA-Nemotron-Nano-9B-v2 model cardHumanity's Last Exam [correct]6.5%fewer than three models have this test yet, or this source ran it for this model alone
$ quantization evidence · 8 measurements

How much of the full-precision result a quantized version kept, measured within one study. Rows from the same study share a table and aren't independent.

precisiontestfull → quantizedkept
NVFP4 PTQ (post-training, W4A4 FP4)IFEval (instruction-level) accuracy (NVIDIA evaluation settings)0.903 → 0.89899.4%source ↗
NVFP4 PTQ (post-training, W4A4 FP4)MATH-500 accuracy (NVIDIA evaluation settings)0.978 → 0.97299.4%source ↗
NVFP4 PTQ (post-training, W4A4 FP4)AIME 2025 accuracy (NVIDIA evaluation settings)0.711 → 0.69898.2%source ↗
NVFP4 PTQ (post-training, W4A4 FP4)GPQA-Diamond accuracy (NVIDIA evaluation settings)0.64 → 0.5992.2%source ↗
NVFP4 QAD (quantization-aware distillation)AIME 2025 accuracy (NVIDIA evaluation settings)0.711 → 0.715100.6%source ↗
NVFP4 QAD (quantization-aware distillation)MATH-500 accuracy (NVIDIA evaluation settings)0.978 → 0.97299.4%source ↗
NVFP4 QAD (quantization-aware distillation)IFEval (instruction-level) accuracy (NVIDIA evaluation settings)0.903 → 0.89398.9%source ↗
NVFP4 QAD (quantization-aware distillation)GPQA-Diamond accuracy (NVIDIA evaluation settings)0.64 → 0.62798.0%source ↗

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json