Nvidia Nemotron 3.5 Lightning

Model Summary

Model Overview

Model Developer: NVIDIA Corporation

Model Dates: December 2025 - May 2026

Data Freshness:

• The pre-training data has a cutoff date of September 2025.
• The post-training data has a cutoff date of May 2026.

What is Nemotron?

NVIDIA Nemotron™ is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents.

Description

NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 is a large language model (LLM) trained by NVIDIA.

The model employs a hybrid Mixture-of-Experts architecture, utilizing interleaved Mamba-2 and MoE layers, along with select Attention layers. The Lightning 3.5 model is released alongside a number of speculative decoding methods for faster text generation. The model has 3B active parameters and 30B parameters in total.

This model is ready for commercial use.

Quick Start

To get quickly started on DGX Spark (GB10) you can use the following command.

Grab the model:

export MODEL_CKPT=nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4
export DSPARK_CKPT=nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark

Run it with vLLM — this recipe uses DSpark speculative decoding, tuned for DGX Spark. (vLLM Nightly: vllm/vllm-openai:v0.27.1)

vllm serve --model $MODEL_CKPT \
  --moe-backend marlin \
  --kv-cache-dtype fp8 \
  --enable-prefix-caching \
  --speculative_config.num_speculative_tokens 3 \
  --mamba-backend flashinfer \
  --mamba-cache-mode align \
  --reasoning-parser nemotron_v3 \
  --speculative_config.model $DSPARK_CKPT \
  --tool-call-parser qwen3_coder \
  --enable-auto-tool-choice

For more details on how to deploy and use the model — see the Quick Start Guide below!

License/Terms of Use

Governing Download Terms: Use of this model is governed by the OpenMDW-1.1 model license.

Benchmarks

Reasoning Benchmark Evaluations

We evaluated our model on the following benchmarks:

Accuracy numbers measured by NVIDIA under a consistent harness (NeMo Gym / Nemo Evaluator SDK); they may differ from vendors' self-reported numbers.

For reproducibility, the evaluation recipes, installation instructions, and commands for NVIDIA Nemotron 3.5 Lightning were collected and published in NeMo Gym. The reported results cover the release evaluation suite, including knowledge and reasoning, instruction following, coding, agentic, tool-use, and long-context. Most evaluations use NeMo Gym-native harnesses while a small subset, including SWE-Bench and Terminal-Bench, used NeMo Evaluator natively. The published recipes specify the benchmark-specific containers, prompts, inference parameters, parser configurations, and scoring settings used to produce the results.

These numbers were measured with and apply to the official NVFP4 checkpoint

Agentic Coding Benchmarks

Additional harness-level coding-agent results for SWE-Bench Verified and Terminal-Bench 2.1 are shown below.

Deployment Geography: Global

Use Case

NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 is a general purpose reasoning and chat model intended to be used in English and coding languages. Other non-English languages (Spanish, French, German, Italian, Japanese) are also supported. Intended for developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. Also suitable for typical instruction-following tasks.

Release Date

Hugging Face — 08/11/2026

Model Architecture

• Architecture Type: Mixture-of-Experts Hybrid (Mamba + Transformer)
• Network Architecture: Nemotron-3-Lightning + Multi-Token Prediction (MTP)
• Number of model parameters: 30B Total / 3B Active

Model Design

The model was pre-trained with over 20T tokens and supports up to 1M context length. The pre-training phase used an NVFP4 recipe. The model includes Multi-Token Prediction (MTP) layers, which predict multiple future tokens to provide richer training signals.

Training Methodology

Stage 1: Pre-Training

• NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 model was pre-trained using an NVFP4 recipe with crawled and synthetic code, math, science, and general knowledge data.
• Software used for pre-training: Megatron-LM

Stage 2: Continued Pre-Training for Multi-Token Prediction (MTP)

• The model underwent a continued pre-training phase to train its Multi-Token Prediction (MTP) layers. In this stage, MTP heads learn to predict multiple future tokens, providing richer training signals to the base model. This phase aligns the MTP layers with the base model's distribution.

Stage 3: Supervised Fine-Tuning

• The model was further fine-tuned on synthetic code, math, science, tool calling, instruction following, structured outputs, and general knowledge data. This stage incorporated data designed to support long-range retrieval and multi-document aggregation.

Stage 4: Reinforcement Learning

• The model underwent multi-environment reinforcement learning using GRPO (Group Relative Policy Optimization) across math, code, science, instruction following, multi-step tool use, multi-turn conversations, and structured output environments. It utilized an asynchronous RL architecture that decouples training from inference and leverages MTP to accelerate rollout generation.
• Software used for reinforcement learning: NeMo RL, NeMo Gym

Stage 5: Post-training Quantization (PTQ)

• We performed post-training quantization (PTQ) with Nvidia Model Optimizer using the following recipe: Four Over Six NVFP4 (a variant of static MSE calibration) W4A16 on routed and shared experts, FP8 per-tensor dynamic scales on mamba in_proj/out_proj and KV cache. We used a subset of the Nemotron Ultra validation set for calibration with 1000 samples at 32k token length.

NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 is a result of the above work.

Input

• Input Type(s): Text
• Input Format(s): String
• Input Parameters: One-Dimensional (1D): Sequences
• Other Properties Related to Input: Maximum context length up to 1M tokens. Supported languages include English, Spanish, French, German, Italian, and Japanese.

Output

• Output Type(s): Text
• Output Format: String
• Output Parameters: One-Dimensional (1D): Sequences
• Other Properties Related to Output: Maximum context length up to 1M tokens

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration

• Runtime Engine(s): PyTorch
• Supported Hardware Microarchitecture Compatibility: NVIDIA Blackwell; NVIDIA Hopper (NVFP4 / W4A16); NVIDIA Ampere (W4A16)
• Preferred/Supported Operating System(s): Linux

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s)

• GA (08/11/2026)

Quick Start Guide

All deployment snippets below assume:

export MODEL_CKPT=nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4

And for DSpark:

export DSPARK_CKPT=nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark

Speculative Decoding Strategies

Lightning 3.5 ships with two external draft models for speculative decoding as well as MTP (Multi-Token Prediction). While we currently recommend DSpark for all cases - your usecase may align with DFlash and MTP:

• DSpark: A semi-autoregressive speculative-decoding drafter that proposes a whole block of candidate tokens in a single forward pass from a parallel backbone. This is recommended for DGX Spark, as well as low-concurrency data centre deployments.
• DFlash: A speculative-decoding drafter that uses a lightweight block-diffusion model to generate an entire draft block in one forward pass.
• MTP: A modeling technique that trains the network to predict several future tokens at each position instead of only the next one.

vLLM

For more indepth instructions on how to deploy through vLLM, head here

• vLLM Nightly: vllm/vllm-openai:v0.27.1

1x DGX Spark (GB10)

Specdec method - DSpark:

vllm serve --model $MODEL_CKPT \
  --moe-backend marlin \
  --kv-cache-dtype fp8 \
  --max-model-len 1048576 \
  --enable-prefix-caching \
  --speculative_config.num_speculative_tokens 3 \
  --mamba-backend flashinfer \
  --mamba-cache-mode align \
  --reasoning-parser nemotron_v3 \
  --speculative_config.method dspark \
  --tool-call-parser qwen3_coder \
  --enable-auto-tool-choice

1x H100

For max throughput deployments, use the following configuration, no speculative decoding strategy is best for this serving configuration, and due to memory constraints the Mamba cache dtype is set as FP16:

vllm serve --model $MODEL_CKPT \
    --max-num-seqs 256 \
    --max-num-batched-tokens 16384 \
    --enable-prefix-caching \
    --async-scheduling \
    --mamba-backend flashinfer \
    --moe-backend humming \
    --linear-backend humming \
    --mamba-ssu-algorithm horizontal \
    --mamba-cache-mode align \
    --mamba-ssm-cache-dtype float16 \
    --enable-mamba-cache-stochastic-rounding \
    --mamba-cache-philox-rounds 5 \
    --reasoning-parser nemotron_v3 \
    --tool-call-parser qwen3_coder \
    --enable-auto-tool-choice

For interactive usage scenarios (achieving 40+ TPS/User) use a lower concurrency (<=128) with DSpark:

vllm serve --model $MODEL_CKPT \
    --max-num-seqs 128 \
    --enable-prefix-caching \
    --async-scheduling \
    --speculative_config.model $DSPARK_CKPT \
    --speculative_config.num_speculative_tokens 3 \
    --mamba-backend flashinfer \
    --mamba-ssm-cache-dtype float16 \
    --enable-mamba-cache-stochastic-rounding \
    --mamba-cache-philox-rounds 5 \
    --reasoning-parser nemotron_v3 \
    --tool-call-parser qwen3_coder \
    --enable-auto-tool-choice

8x H100

For long-context, multi-GPU serving (TP8 with expert parallelism):

vllm serve --model $MODEL_CKPT \
    --mamba-backend flashinfer \
    --async-scheduling \
    --enable-prefix-caching \
    --mamba-cache-mode align \
    --enable-expert-parallel \
    --tensor-parallel-size 8 \
    --reasoning-parser nemotron_v3 \
    --tool-call-parser qwen3_coder \
    --enable-auto-tool-choice

1x GB200

vllm serve --model $MODEL_CKPT \
    --max-num-batched-tokens 10240 \
    --no-enable-prefix-caching \
    --async-scheduling \
    --speculative_config.model $DSPARK_CKPT \
    --speculative_config.num_speculative_tokens 5 \
    --mamba-backend flashinfer \
    --reasoning-parser nemotron_v3 \
    --tool-call-parser qwen3_coder \
    --enable-auto-tool-choice

W4A16 — Ampere

The same checkpoint also serves via W4A16 kernels, extending coverage to Ampere-class GPUs:

vllm serve --model nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 \
    --moe-backend humming \
    --linear-backend humming \
    --max-num-seqs 256 \
    --max-num-batched-tokens 32768 \
    --enable-prefix-caching \
    --async-scheduling \
    --quantization modelopt_fp4 \
    --mamba-backend flashinfer \
    --mamba-cache-mode align \
    --mamba-ssu-algorithm simple \
    --reasoning-parser nemotron_v3 \
    --tool-call-parser qwen3_coder \
    --enable-auto-tool-choice

• Context Length: The H100 and GB200 snippets above serve the model's full 1M-token context window by default. If you're memory-constrained — or want more KV-cache headroom at high concurrency — lower --max-model-len to match your workload.

TensorRT-LLM

For more indepth instructions on how to deploy through TensorRT-LLM, head here

Container: nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc24

1x H100

cat > nemotron-35-lightning-nvfp4-mtp.yaml << EOF
kv_cache_config:
  dtype: fp8
  enable_block_reuse: false
  mamba_state_config:
     periodic_snapshot_interval: 8192
  free_gpu_memory_fraction: 0.8
  mamba_ssm_cache_dtype: float16
  mamba_ssm_stochastic_rounding: true
  mamba_ssm_philox_rounds: 5
moe_config:
   backend: MARLIN
nvfp4_gemm_config:
  allowed_backends: [marlin, cutlass, cublaslt, cuda_core]
cuda_graph_config:
    enable_padding: true
    max_batch_size: 8
speculative_config:
  decoding_type: MTP
  max_draft_len: 3
  allow_advanced_sampling: true
enable_chunked_prefill: true
num_postprocess_workers: 4
print_iter_log: true
stream_interval: 10
disable_overlap_scheduler: false
EOF

trtllm-serve \
$MODEL_CKPT \
--max_batch_size 8 \
--max_num_tokens 8192 \
--reasoning_parser nemotron-v3 \
--tool_parser qwen3_coder \
--config nemotron-35-lightning-nvfp4-mtp.yaml

• Context length: The command above serves the model's full 1M-token context window by default. If you're memory-constrained — or want more KV-cache headroom at higher concurrency — lower --max_seq_len to match your workload.

SGLang

For more indepth instructions on how to deploy through SGLang, head here

• Container: lmsysorg/sglang:dev-nemotron3-5-lightning

1x H100

sglang serve \
    --model-path nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 \
    --max-running-requests 256 \
    --trust-remote-code \
    --chunked-prefill-size 32768 \
    --mem-fraction-static 0.9 \
    --speculative-algorithm EAGLE \
    --speculative-draft-model-path nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 \
    --speculative-num-steps 3 \
    --speculative-eagle-topk 1 \
    --speculative-num-draft-tokens 4 \
    --mamba-backend flashinfer \
    --mamba-radix-cache-strategy extra_buffer \
    --reasoning-parser nemotron_3 \
    --tool-call-parser qwen3_coder

• Context length: The command above serves the model's full 1M-token context window by default. If you're memory-constrained — or want more KV-cache headroom at higher concurrency — set --context-length to a smaller value.

API Client

The examples below use the OpenAI-compatible client and work with any of the serving backends above. All backends serve on port 8000 (vLLM and TRT-LLM by default; SGLang via $PORT=8000), so the base_url works as-is. Recommended sampling settings are Temperature 1.0 and Top_P 0.95.

The vLLM snippets above register the model as nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 via --served-model-name. For the other backends — or if you change that flag — copy the identifier returned by GET /v1/models into MODEL below.

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
MODEL = "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4"

Lightning 3.5 exposes reasoning control through chat-template kwargs: thinking enabled (the default), and thinking disabled for direct answers.

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "Write a haiku about GPUs"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={"chat_template_kwargs": {"enable_thinking": True}}
)
print(response.choices[0].message.content)

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "What is the capital of Japan?"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={"chat_template_kwargs": {"enable_thinking": False}}
)
print(response.choices[0].message.content)

stream = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "Explain speculative decoding in two sentences"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Tool Calling

The TRT-LLM snippet above already launches with the required parsers (--reasoning_parser nemotron-v3 --tool_parser qwen3_coder). For vLLM, add the following to any serve command above:

    --enable-auto-tool-choice \
    --tool-call-parser qwen3_coder \
    --reasoning-parser nemotron_v3

NOTE: For coding agents, add extra_body={"chat_template_kwargs": {"force_nonempty_content": True}} to the API call, as shown below.

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "What's the weather in Santa Clara?"}],
    tools=tools,
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={"chat_template_kwargs": {"force_nonempty_content": True}},
)
print(response.choices[0].message.tool_calls)

Training, Testing, and Evaluation Datasets

Data Modality: Text Training Data Size: More than 20 Trillion Tokens Dataset partition: Training [100%], testing [0%], validation [0%] Time period for training data collection: 2013 to December 2025 Time period for testing data collection: 2013 to December 2025 Time period for validation data collection: 2013 to December 2025 Data Collection Method by dataset: Hybrid: Automated, Manually-Collected, Synthetic Labeling Method by dataset: Hybrid: Automated, Manually-Labeled, Synthetic

NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 is pre-trained on a large corpus of high-quality curated and synthetically-generated data. It is trained in the English language, as well as 19 other spoken languages and 43 programming languages. Our sources cover a variety of document types such as: webpages, dialogue, articles, and other written materials. The corpus spans domains including legal, math, science, finance, and more. We also include a small portion of question-answering, and alignment style data to improve model accuracy. The model was pre-trained for more than 20 trillion tokens.

The post-training corpus for NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 consists of high-quality curated and synthetically-generated data. Primary languages used for post-training include English, French, German, Italian, Japanese, Spanish, and Chinese.

These datasets, such as FinePDFs, EssentialWeb, HotpotQA, SQuAD, and HelpSteer3, do not collectively or exhaustively represent all demographic groups (and proportionally therein). For instance, these datasets do not contain explicit mentions of demographic classes such as age, gender, or ethnicity in 64-99% of samples, depending on the source. In the subset where such terms are present, document-based datasets (FinePDFs and EssentialWeb) contain representational skews, such as references to "male" outnumbering those to "female", and mentions of "White" as the most frequent among ethnic identifiers (comprising 43-44% of ethnicity mentions). To mitigate these imbalances, we recommend considering evaluation techniques such as bias audits, fine-tuning with demographically balanced datasets, and mitigation strategies like counterfactual data augmentation to align with the desired model behavior. This evaluation used a 3,000-sample subset per dataset, identified as the optimal threshold for maximizing embedder accuracy.

During post-training, we generate synthetic data by distilling trajectories, solutions, and translations from strong teacher models and agent systems, often grounded in real tasks or documents and aggressively filtered for quality. For math, code, and science, we start from curated problem sets and use open source permissive models such as GPT-OSS-120B to produce step-by-step reasoning traces, candidate solutions, best-of-n selection traces, and verified CUDA kernels. For long-context and science, we build synthetic QA and reasoning data by retrieving passages from long documents, generating MCQ/OpenQA questions and answers, and paraphrasing them into multiple prompt/response formats to ensure diversity. Across all pipelines we stack automated verification—compilers, numerical checks, language identification—to ensure our data is high quality.

For all domains, we apply a unified data filtering pipeline to ensure that only high-quality, license-compliant, and verifiable samples are used for post-training. We first discard malformed examples using structural checks (e.g., missing tool definitions when tool calls are present). We then aggressively filter reasoning traces exhibiting pathological repetition, such as repeated n-grams within a sliding window or across the entire trajectory, which we found to be a strong indicator of malformed or low-quality reasoning. Finally, based on internal audits of synthetically generated datasets, we observed that some teacher models occasionally produce reasoning traces and final responses that implicitly align with specific political entities or promote nationalistic narratives. To mitigate this, we apply targeted keyword- and regex-based filters and remove all trajectories matching such behavior.

Alongside the model, we release our final pre-training and post-training data, as outlined in this section. For ease of analysis, there is a sample set that is ungated. For all remaining code, math and multilingual data, gating and approval is required, and the dataset is permissively licensed for model training purposes.

Testing Datasets:

Data Collection Method by dataset

• Hybrid: Automated, Manually-Collected, Synthetic Labeling Method by dataset
  
• Hybrid: Automated, Manually-Labeled, Synthetic Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI. These benchmarks test model capabilities on tasks such as tool-calling and instruction following.

Evaluation Datasets:

Data Collection Method by dataset

• Hybrid: Automated, Manually-Collected, Synthetic Labeling Method by dataset
  
• Hybrid: Automated, Manually-Labeled, Synthetic Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI. These benchmarks test model capabilities on tasks such as tool-calling and instruction following.

Inference

• Acceleration Engine: PyTorch
• Test Hardware:
  • NVIDIA Hopper
    • 1-8x H100
    • 1-8x H200
  • NVIDIA Blackwell
    • GB200
    • DGX Spark (GB10)
    • GeForce RTX 5090

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

We advise against circumvention of any provided safety guardrails contained in the Model without a substantially similar guardrail appropriate for your use case. For more details: Safety and Explainability Subcards.

For more detailed information on ethical considerations for this model, please see the Model Card++ Bias, and Privacy Subcards.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.