The post-training data has a cutoff date of February 2026.
The pre-training data has a cutoff date of June 2025.
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
Nemotron-3-Super-120B-A12B-BF16 is a large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities. It is optimized for collaborative agents and high-volume workloads such as IT ticket automation. Like other models in the family, it responds to user queries and tasks by first generating a reasoning trace and then concluding with a final response. The model's reasoning capabilities can be configured through a flag in the chat template.
The model employs a hybrid Latent Mixture-of-Experts (LatentMoE) architecture, utilizing interleaved Mamba-2 and MoE layers, along with select Attention layers. Distinct from the Nano model, the Super model incorporates Multi-Token Prediction (MTP) layers for faster text generation and improved quality, and it is trained using NVFP4 quantization to maximize compute efficiency. The model has 12B active parameters and 120B parameters in total.
The supported languages include: English, French, German, Italian, Japanese, Spanish, and Chinese
All evaluation results were collected via Nemo Evaluator SDK and for most benchmarks, the Nemo Skills Harness. For reproducibility purposes, more details on the evaluation settings can be found in the Nemo Evaluator SDK configs folder and the reproducibility tutorial for Nemotron 3 Super. The open source container on Nemo Skills packaged via NVIDIA's Nemo Evaluator SDK used for evaluations can be found here. In addition to Nemo Skills, the evaluations also used dedicated open-source packaged containers for Tau-2 Bench (default prompt), Terminal Bench Hard (48 tasks), ScaleAI Multi Challenge Multi-turn Instruction Following, and Ruler.
The following benchmarks are not onboarded yet in our open source tools and for these we used either their official open source implementation or otherwise an internal scaffolding that we plan to open source in the future: SWE Bench Verified (OpenHands), SWE Bench Multilingual (OpenHands), BrowseComp with Search (internal implementation with Serp API), Terminal Bench Core 2.0 (Harbor).
Deployment Geography: Global
Use Case
NVIDIA-Nemotron-3-Super-120B-A12B-BF16 is a general purpose reasoning and chat model intended to be used in English, Code, and supported multilingual contexts. This model is optimized for collaborative agents and high-volume workloads. It is intended to be used by developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. This model is also suitable for complex instruction-following tasks and long-context reasoning.
Architecture Type: Mamba2-Transformer Hybrid Latent Mixture of Experts (LatentMoE) with Multi-Token Prediction (MTP)
Network Architecture: Nemotron Hybrid LatentMoE
Number of model parameters: 120B Total / 12B Active
Model Design
The model utilizes the LatentMoE architecture, where tokens are projected into a smaller latent dimension for expert routing and computation, improving accuracy per byte. The Super model is pre-trained using NVFP4 quantization — the first model in the Nemotron 3 family trained at this precision. The majority of linear layers use NVFP4 for weights, activations, and gradients, while select layers (including latent projections, MTP layers, QKV/attention projections, and embeddings) are maintained in BF16 or MXFP8 for training stability. The model includes Multi-Token Prediction (MTP) layers using a shared-weight design across prediction heads. This improves training signal quality, enables faster inference via native speculative decoding, and supports more stable autoregressive drafting at longer draft lengths compared to independently trained offset heads.
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. All datasets are disclosed in the Training and Evaluation Datasets section of this document. Major portions of the fine-tuning corpus are released in the Nemotron-Post-Training-v3 collection. Data Designer is one of the libraries used to prepare these corpora.
Stage 3: Reinforcement Learning
The model underwent multi-environment reinforcement learning using asynchronous 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 fully decouples training from inference across separate GPU devices, leveraging in-flight weight updates and MTP to accelerate rollout generation. Conversational quality was further refined through RLHF. All datasets are disclosed in the Training and Evaluation Datasets section of this document. The RL environments and datasets are released as part of NeMo Gym.
Other Properties Related to Input: Maximum context length up to 1M tokens. Supported languages include: English, French, German, Italian, Japanese, Spanish, and Chinese
Other Properties Related to Output: Maximum context length up to 1M tokens
Our AI models are designed and 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.
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)
v1.0 - GA
Quick Start Guide
For each inference backend, you'll need the custom super_v3 reasoning parser. Download it with:
Context length defaults to 256k above. To use up to 1M, set VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 and --max-model-len 1048576.
B200/B300 (BF16): The larger HBM capacity per device means the BF16 checkpoint fits on 2 GPUs. Set --tensor-parallel-size 2 and remove --enable-expert-parallel. All other flags remain the same.
B200/B300 (BF16): The larger HBM capacity per device means the BF16 checkpoint fits on 2 GPUs. Set --tp_size 2 --ep_size 2 and reduce max_batch_size to 128 in both the config file and the serve command. All other flags remain the same.
API Client
The examples below use the OpenAI-compatible client and work with any of the serving backends above.
NOTE: For coding agents add the following to the API call - extra_body={“chat_template_kwargs”: {“force_nonempty_content”: True}
1response = client.chat.completions.create(2 model=MODEL,3 messages=[{"role":"user","content":"Write a haiku about GPUs"}],4 max_tokens=16000,5 temperature=1.0,6 top_p=0.95,7 extra_body={"chat_template_kwargs":{"enable_thinking":True}}8)9print(response.choices[0].message.content)
Reasoning OFF
python
1response = client.chat.completions.create(2 model=MODEL,3 messages=[{"role":"user","content":"What is the capital of Japan?"}],4 max_tokens=16000,5 temperature=1.0,6 top_p=0.95,7 extra_body={"chat_template_kwargs":{"enable_thinking":False}}8)9print(response.choices[0].message.content)
Low-effort reasoning
Uses significantly fewer reasoning tokens than full thinking mode. Recommended as a starting point before tuning explicit token budgets.
python
1response = client.chat.completions.create(2 model=MODEL,3 messages=[{"role":"user","content":"What is the capital of Japan?"}],4 max_tokens=16000,5 temperature=1.0,6 top_p=0.95,7 extra_body={"chat_template_kwargs":{"enable_thinking":True,"low_effort":True}}8)9print(response.choices[0].message.content)
OpenCode
OpenCode is an AI coding agent that runs in your terminal. It connects to any OpenAI-compatible endpoint, making it compatible with all three serving backends above (vLLM, SGLang, and TRT-LLM).
Create or update your ~/.config/opencode/opencode.json:
Update baseURL to match whichever backend you are running. The default port above (8000) matches the vLLM example; SGLang and TRT-LLM use 30000 and 8123 respectively.
To learn more about other supported agent scaffolds - check out this resource
Advanced: Budget-Controlled Reasoning
Set a hard token ceiling on the reasoning trace using reasoning_budget. The model will attempt to close the trace at the next newline before the budget is hit; if none is found within 500 tokens it closes abruptly at reasoning_budget + 500.
python
1from typing import Any, Dict, List
2import openai
3from transformers import AutoTokenizer
456classThinkingBudgetClient:7def__init__(self, base_url:str, api_key:str, tokenizer_name_or_path:str):8 self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_name_or_path)9 self.client = openai.OpenAI(base_url=base_url, api_key=api_key)1011defchat_completion(12 self,13 model:str,14 messages: List[Dict[str, Any]],15 reasoning_budget:int=512,16 max_tokens:int=1024,17**kwargs,18)-> Dict[str, Any]:19assert max_tokens > reasoning_budget,(20f"reasoning_budget must be less than max_tokens. "21f"Got {max_tokens=} and {reasoning_budget=}"22)2324# Step 1: generate the reasoning trace up to the budget25 response = self.client.chat.completions.create(26 model=model, messages=messages, max_tokens=reasoning_budget,**kwargs
27)28 reasoning_content = response.choices[0].message.content
29if""notin reasoning_content:30 reasoning_content =f"{reasoning_content}.\n\n\n"3132 reasoning_tokens_len =len(33 self.tokenizer.encode(reasoning_content, add_special_tokens=False)34)35 remaining_tokens = max_tokens - reasoning_tokens_len
36assert remaining_tokens >0,(37f"No tokens remaining for response ({remaining_tokens=}). "38"Increase max_tokens or lower reasoning_budget."39)4041# Step 2: continue from the closed reasoning trace42 messages.append({"role":"assistant","content": reasoning_content})43 prompt = self.tokenizer.apply_chat_template(44 messages, tokenize=False, continue_final_message=True45)46 response = self.client.completions.create(47 model=model, prompt=prompt, max_tokens=remaining_tokens,**kwargs
48)4950return{51"reasoning_content": reasoning_content.strip().strip("").strip(),52"content": response.choices[0].text,53"finish_reason": response.choices[0].finish_reason,54}
Example usage (32-token reasoning budget):
python
1client = ThinkingBudgetClient(2 base_url="http://localhost:8000/v1",3 api_key="EMPTY",4 tokenizer_name_or_path="nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16",5)67result = client.chat_completion(8 model="nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16",9 messages=[10{"role":"system","content":"You are a helpful assistant. /think"},11{"role":"user","content":"What is 2+2?"},12],13 reasoning_budget=32,14 max_tokens=512,15 temperature=1.0,16 top_p=0.95,17)18print(result)
Transformers
The model has been integrated into 🤗 Transformers since v5.3.0. We recommend using the Nemotron 3 Super container from the NeMo Framework to ensure all required libraries are available.
Please note that the model supports up to a 1M context size, although the default context size in the Hugging Face configuration is 256k due to higher VRAM requirements.
Here is an example of generating outputs with reasoning enabled (the default):
Data Modality: Text
The total size: 15,573,172,908,990 Tokens
Total number of datasets: 153
Dataset partition:Training [100%], testing [0%], validation [0%]Time period for training data collection: 2013 to February 24, 2026
Time period for testing data collection: 2013 to February 24, 2026
Time period for validation data collection: 2013 to February 24, 2026
Data Collection Method by dataset: Hybrid: Automated, Human, Synthetic
Labeling Method by dataset: Hybrid: Automated, Human, Synthetic
NVIDIA-Nemotron-3-Super-120B-A12B-BF16 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 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 trained for approximately 25 trillion tokens.
The post-training corpus for NVIDIA-Nemotron-3-Super-120B-A12B-BF16 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.
More details on the datasets and synthetic data generation methods can be found in the technical report NVIDIA Nemotron 3 Super.
Click to explore the full dataset catalogue used for training
Base Pre-Training Corpus (Nemotron 3 Foundation)
The foundation of the model is trained on the Nemotron-3-Nano corpus, comprising the following collections:
Dataset Collection
Token Counts
Description
Nemotron-CC-v2 & v2.1
9.13T
A massive collection of English web data filtered from Common Crawl, including 2.5T+ tokens of new organic, translated, and synthetically rephrased content.
Nemotron-CC-Code-v1
427.9B
High-quality code tokens extracted from Common Crawl using the Lynx + LLM pipeline to preserve structure and equations.
Nemotron-Pretraining-Code-v1 & v2
1.09T
Curated GitHub code references with multi-stage filtering, deduplication, and large-scale synthetic code data.
Nemotron-CC-Math-v1
133.3B
High-quality math pre-training dataset preserving LaTeX formatting and mathematical structures.
Nemotron-Pretraining-Specialized-v1
336.4B
Synthetic datasets targeting specialized domains such as STEM reasoning and scientific coding.
The English Common Crawl data was downloaded from the Common Crawl Foundation (see their FAQ for details on their crawling) and includes the snapshots CC-MAIN-2013-20 through CC-MAIN-2025-13. The data was subsequently deduplicated and filtered in various ways described in the Nemotron-CC paper. Additionally, we extracted data for fifteen languages from the following three Common Crawl snapshots: CC-MAIN-2024-51, CC-MAIN-2025-08, CC-MAIN-2025-18. The fifteen languages included were Arabic, Chinese, Danish, Dutch, French, German, Italian, Japanese, Korean, Polish, Portuguese, Russian, Spanish, Swedish, and Thai. As we did not have reliable multilingual model-based quality classifiers available, we applied just heuristic filtering instead—similar to what we did for lower quality English data in the Nemotron-CC pipeline, but selectively removing some filters for some languages that did not work well. Deduplication was done in the same way as for Nemotron-CC.
The GitHub Crawl was collected using the GitHub REST API and the Amazon S3 API. Each crawl was operated in accordance with the rate limits set by its respective source, either GitHub or S3. We collect raw source code and subsequently remove any having a license which does not exist in our permissive-license set (for additional details, refer to the technical report).
Dataset
Modality
Dataset Size
Collection Period
Collecting Organisation
English Common Crawl
Text
3.36T
4/8/2025
NVIDIA Advanced Deep Learning Research
English Common Crawl 1.1
Text
Not disclosed
10/2/2025
NVIDIA Advanced Deep Learning Research
Multilingual Common Crawl
Text
812.7B
5/1/2025
NVIDIA Advanced Deep Learning Research
GitHub Crawl
Text
747.4B
4/29/2025
NVIDIA Advanced Deep Learning Research
Private Non-publicly Accessible Datasets of Third Parties
Dataset
Model(s) used
Global Regulation
Unknown
TAUS Translation Memory
Unknown
Scale HLE
Unknown
HackerRank Coding
Unknown
RL data for Search
Gemini 3; GPT-5 *
Models used for prompt generation only
Private Non-publicly Accessible Datasets by NVIDIA
Dataset
Model(s) used
Simple Minesweeper
-
Simple Sudoku
-
Multitool Typewriter Hard
-
Machine Translation of News Commentary and TAUS Translation Memory
Synthetic Multilingual Science and Code data from DeepSeek-R1, DeepSeek-R1-0528, Qwen2.5-32B-Instruct, and Qwen3-235B-A22B, translated with Qwen2.5-32B-Instruct and Qwen2.5-14B-Instruct
Synthetic Structured Outputs from Qwen3-30B-A3B-Instruct-2507, Qwen3-30B-A3B-Thinking-2507, Qwen3-235B-A22B-Instruct-2507, and Qwen3-235B-A22B-Thinking-2507
For our post-training recipe, we focused on 9 main languages in addition to English: French, German, Italian, Japanese, Spanish, and Chinese
Those languages were represented in the form of multilingual reasoning and translation tasks.
The following table depicts our sample distribution for the 6 languages and 5 translation pairs.
Language
Size
English
13.48M
Italian
53k
German
53k
Spanish
53k
French
53k
Japanese
53k
Chinese
53k
English <-> Italian
43.2k
English <-> German
43.2k
English <-> Spanish
43.2k
English <-> French
43.2k
English <-> Japanese
43.2k
Evaluation Dataset
Data Collection Method by dataset: Hybrid: Human, Synthetic
Labeling Method by dataset: Hybrid: Automated, Human, Synthetic
Inference
Acceleration Engine: PyTorch
Test Hardware:
NVIDIA Hopper
1-8x H100
1-8x H200
NVIDIA Grace Blackwell
GB200
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. When downloaded or used in accordance with our terms of service, 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.
Citation
bibtex
1@misc{nvidia_nemotron_3_2025,
2 title = {NVIDIA Nemotron 3: Efficient and Open Intelligence},
3 author = {{NVIDIA}},
4 year = {2025},
5 url = {https://arxiv.org/abs/2512.20856},
6 note = {White Paper}
7}
