Skip to main content
Watch: I Built an AI App Without Leaving Databricks

NVIDIA - Mixture of experts - verified

Run NVIDIA Nemotron 3.5 Lightning 30B-A3B locally

NVIDIA Nemotron 3.5 Lightning 30B-A3B is a 31.6B mixture-of-experts (3B active) model from NVIDIA with a 1.05M-token context window, released under OpenMDW License 1.1 (commercial use is allowed); the smallest listed download is 17.8 GB, the 4bit MLX build from mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit.

Official weights: nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

Total params
31.6B
Active params
3B
Context
1.05M
Smallest download
17.8 GB

Fits in

Smallest download at most 85% of memory

Downloads

The official weights and every quantization listed for this model, with file sizes from Hugging Face.

NVIDIA Nemotron 3.5 Lightning 30B-A3B downloads and file sizes
DownloadFormatSizeRepo
Official weightsBF1665.8 GBnvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
Q4_0GGUF18.9 GBggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF
Q8_0GGUF33.6 GBggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF
4bitSmallest listedMLX17.8 GBmlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit

GB is decimal: 1 GB = 10^9 bytes. Add memory for the context and the operating system on top of the file size.

Run it

Copied verbatim from the page linked under each command. Server commands from model cards often assume a multi-GPU node; check the flags before you run them.

Ollama

Terminal
ollama run nemotron-3.5-lightning

Source: ollama.com/library/nemotron-3.5-lightning

vLLM

Terminal
vllm serve --model $MODEL_CKPT \
    --max-num-seqs 128 \
    --enable-prefix-caching \
    --async-scheduling \
    --mamba-backend flashinfer \
    --mamba-ssm-cache-dtype float16 \
    --enable-mamba-cache-stochastic-rounding \
    --mamba-cache-philox-rounds 5

Source: huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

SGLang

Terminal
sglang serve \
    --model-path $MODEL_CKPT \
    --mamba-ssm-dtype float16 \
    --mem-fraction-static 0.85 \
    --cuda-graph-max-bs-decode 16 \
    --reasoning-parser nemotron_3 \
    --tool-call-parser qwen3_coder \
    --port 8000

Source: huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

Not sure which runtime to use? Compare local LLM runtimes by platform, API and features.

License

Commercial use
Commercial use: YesCommercial use is allowed.
Gated download
No. The files download directly.

Worth knowing

Hybrid Mamba-2 and MoE model; the card's vLLM and SGLang commands expect MODEL_CKPT exported as the repo id first, and NVIDIA points to its NVFP4 release for optimized inference.

Released (per the model card).

FAQ

How big is the NVIDIA Nemotron 3.5 Lightning 30B-A3B download?

The official BF16 weights in nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 are 65.8 GB. The smallest listed download is 17.8 GB, the 4bit MLX build from mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit. Plan for more memory than the file size: the context (KV cache) and the operating system need room too.

Can I use NVIDIA Nemotron 3.5 Lightning 30B-A3B commercially?

NVIDIA Nemotron 3.5 Lightning 30B-A3B is released under OpenMDW License 1.1, and commercial use is allowed. The license text is at https://openmdw.ai/license/1-1/.

How much memory do I need for NVIDIA Nemotron 3.5 Lightning 30B-A3B?

By this directory's rule of thumb it fits in 24 GB and up: the smallest listed download, 17.8 GB, is at most 85% of 24 GB, which leaves room for the context and the operating system.

How do I run NVIDIA Nemotron 3.5 Lightning 30B-A3B with Ollama?

Run "ollama run nemotron-3.5-lightning", as shown at https://ollama.com/library/nemotron-3.5-lightning. The Ollama library lists the nemotron-3.5-lightning:30b tag at 19GB - 25GB.

What is the context length of NVIDIA Nemotron 3.5 Lightning 30B-A3B?

1,048,576 tokens. A longer context needs more memory for the KV cache, so you can run with a shorter one when memory is tight.

Sources

Last verified from the pages below.

Newsletter

New open weights, every week

One email per week. New model releases, open-weight drops and what they mean for what you build.

Keep exploring

Best local AI models

Our curated picks by hardware tier, with benchmarks and the reasoning.

Model API pricing and specs

First-party prices per 1M tokens when you would rather call an API.

AI benchmark leaderboards

Sourced scores, for when you need to compare quality.

Setup guides

Point your coding tools at the model you picked.