For more details on how to deploy and use the model - see the Quick Start Guide below!
Model Overview
Model Developer: NVIDIA Corporation
Model Dates: December 2025 - April 2026
Data Freshness:
The post-training data has a cutoff date of May 2026.
The pre-training data has a cutoff date of September 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-Ultra-550B-A55B-NVFP4 is a frontier-scale large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities. It is optimized for the most demanding workloads, including complex multi-step agents, long-context analysis, and high-accuracy reasoning over code, math, and science. 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. Like the Super model, the Ultra model incorporates Multi-Token Prediction (MTP) layers for faster text generation and improved quality, and it is trained using an NVFP4 pre-training recipe to maximize compute efficiency. The model has 55B active parameters and 550B parameters in total.
The supported languages include: English, French, Spanish, Italian, German, Japanese, Korean, Hindi, Brazilian Portuguese, and Chinese.
This model is ready for commercial and non-commercial use.
All evaluation results were collected via Nemo Evaluator SDK. We used three main evaluation harnesses: Nemo Gym, Nemo Skills, and Harbor with extended sandboxing support via AWS ECS on Nemo Evaluator. In addition, the evaluations also used dedicated open-source packaged containers for ScaleAI Multi Challenge Multi Turn Instruction Following and KernelBench. For reproducibility purposes, more details on the evaluation settings and pinned containers can be found in the Nemo Evaluator SDK examples folder and the reproducibility tutorial for Nemotron 3 Ultra.
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: BrowseComp with Search, Tau Bench 3, ProfBench with Search, PinchBench, Vals.ai, LongBench v2.
Deployment Geography: Global
Use Case
NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 is a frontier-scale general purpose reasoning and chat model intended to be used in English, Code, and supported multilingual contexts. This model is optimized for complex agentic workflows, long-context reasoning, and high-stakes analytical 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 over very large documents and codebases.
Architecture Type: Mamba2-Transformer Hybrid Latent Mixture of Experts (LatentMoE) with Multi-Token Prediction (MTP)
Network Architecture: Nemotron Hybrid LatentMoE
Number of model parameters: 550B Total / 55B 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 Ultra model is pre-trained using an NVFP4 recipe — sharing the quantization-aware pre-training approach pioneered in the Nemotron 3 family. 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.
Training Methodology
Stage 1: Pre-Training
NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 model was pre-trained for approximately 20T tokens using crawled and synthetic code, math, science, and general knowledge data. Training leveraged an NVFP4 recipe for efficiency. All datasets are disclosed in the Training and Evaluation Datasets section of this document. Major portions of the pre-training corpus are released in the Nemotron-Pre-Training-Datasets collection.
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.
The model underwent Multi-Domain On-Policy Distillation (MOPD) to improve reasoning across many task types while staying efficient. This technique uses strong teacher models to guide training on the model's own generated attempts (on-policy rollouts), helping recover accuracy and improve performance across coding, math, instruction following, tool use, and agentic workflows. By distilling teacher signal onto the student's own trajectories rather than offline traces, MOPD better aligns the student's behavior with what it would actually produce at inference time, yielding stronger gains than purely off-policy distillation.
NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 model is a result of the above work.
Other Properties Related to Input: Maximum context length up to 1M tokens. Supported languages include: English, French, Spanish, Italian, German, Japanese, Hindi, Korean, Brazilian Portuguese, 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
The Ultra NVFP4 checkpoint is a frontier-scale model quantized for maximum throughput on the latest hardware. The minimum recommended hardware is:
Single-node: 4× B200 (fits NVFP4 weights plus KV cache with headroom)
Multi-node: ≥4 GPUs across GB200 / GB300
All deployment snippets below default to port 8000, with chunked prefill, NVFP4 KV caching, and MTP (5 speculative tokens) enabled.
Multi-Node Setup with Ray (Recommended for multi-node deployments)
The recommended multi-processing backend for multi-node deployments is Ray v2. Below is a template for launching a Ray cluster:
bash
1# Set the IP for the head node in RAY_HEAD_IP2exportRAY_HEAD_IP=<head_node_ip>3exportRAY_PORT=63794exportRAY_ADDRESS=${RAY_HEAD_IP}:${RAY_PORT}56# Start Ray head node (vLLM/SGLang will run on this node)7ray start --head --node-ip-address=${RAY_HEAD_IP} --port=${RAY_PORT}89# Start Ray worker node(s)10ray start --address=${RAY_HEAD_IP}:${RAY_PORT} --block
1112# Verify Ray cluster is ready13ray status --address=${RAY_HEAD_IP}:${RAY_PORT}
Note:ray[cgraph] is required: uv pip install "ray[cgraph]"
vLLM
Recommended container:vllm/vllm-openai:v0.22.0
For more detailed information, please see this cookbook.
Context length: Defaults to 256k above. To use up to 1M, set SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 and --context-length 1048576.
Tool calls + reasoning parsing: When calling the chat completions endpoint with tools, you must set "chat_template_kwargs": {"enable_thinking": true, "force_nonempty_content": true} in the request body to parse both reasoning and tool calls correctly.
Long-context configuration: For long-context benchmarking, set TLLM_ALLOW_LONG_MAX_MODEL_LEN=1 as an environment variable and add --max_seq_len <seq_len> as the desired maximum context length. The MTP speculative_config block above carries over unchanged — on rc16, max_draft_len is the authoritative field and num_nextn_predict_layers is treated as deprecated.
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.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.message.content)
Medium-effort reasoning
Uses significantly fewer reasoning tokens than full thinking mode. Recommended as a starting point before tuning explicit token budgets.
py
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,"medium_effort":True}}8)9print(response.choices[0].message.content)
Tool calling with reasoning (SGLang requires explicit chat template kwargs)
python
1response = client.chat.completions.create(2 model=MODEL,3 messages=[{"role":"user","content":"What's the weather in New York?"}],4 tools=[{5"type":"function",6"function":{7"name":"get_weather",8"description":"Get the current weather for a city.",9"parameters":{10"type":"object",11"properties":{12"city":{"type":"string"},13"unit":{"type":"string","enum":["celsius","fahrenheit"]}14},15"required":["city"]16}17}18}],19 tool_choice="required",20 max_tokens=256,21 temperature=1.0,22 top_p=0.95,23 extra_body={"chat_template_kwargs":{"enable_thinking":True,"force_nonempty_content":True}}24)
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:
All backends above default to port 8000, so the baseURL works as-is for vLLM, SGLang, and TRT-LLM. To learn more about other supported agent scaffolds, check out this resource.
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.message.content
29if"</think>"notin reasoning_content:30 reasoning_content =f"{reasoning_content}.\n\n</think>\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("</think>").strip(),52"content": response.choices.text,53"finish_reason": response.choices.finish_reason,54}5556# Example usage (32-token reasoning budget):57client = ThinkingBudgetClient(58 base_url="http://localhost:8000/v1",59 api_key="EMPTY",60 tokenizer_name_or_path="nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4",61)6263result = client.chat_completion(64 model="nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4",65 messages=[66{"role":"system","content":"You are a helpful assistant. /think"},67{"role":"user","content":"What is 2+2?"},68],69 reasoning_budget=32,70 max_tokens=512,71 temperature=1.0,72 top_p=0.95,73)74print(result)
Training and Evaluation Datasets
Training
Data Modality: Text The total size: 53.8 TiB (14.8 trillion tokens) Total number of datasets: 226 Dataset partition:Training [100%], testing [0%], validation [0%] Time period for training data collection: 2013 to 2026 Time period for testing data collection: 2013 to 2026 Time period for validation data collection: 2013 to 2026 Data Collection Method by dataset: Hybrid: Automated, Human, Synthetic Labeling Method by dataset: Hybrid: Automated, Human, Synthetic
NVIDIA-Nemotron-3-Ultra-550B-A55B-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 11 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 pre-trained for approximately 20 trillion tokens.
The post-training corpus for NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 consists of high-quality curated and synthetically-generated data. Primary languages used for post-training include English, French, Spanish, Italian, German, Japanese, Hindi, Korean, Brazilian Portuguese, 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 Ultra.
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 & v3
1.7T
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.
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
GitHub Crawl 1.1
Text
172.7B
9/30/2025
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
Private Non-publicly Accessible Datasets by NVIDIA
Dataset
Model(s) used
Simple Minesweeper
Undisclosed
Simple Sudoku
Undisclosed
Multitool Typewriter Hard
Undisclosed
Machine Translation of News Commentary and TAUS Translation Memory
CommonPile (caselaw_access_project_filtered); California Code of Regulations; Judicial Ethics Opinions; GLOBALCIT; CUAD; Nemotron Personas; ToSDR Terms of Service Corpus; CodeHima/TOS_Dataset; ContractNLI; CaseHOLD; Code of Federal Regulations; Canadian Case Law (subsets that allow commercial use)
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
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[Qwen2.5-32B]
Synthetic Abstention Data from Nemotron Super v3
Text
Undisclosed
[Go abstention Dataset]
[nvidia/nvidia/nemotron-3-super-v3]
Synthetic Chemistry Data from Nemotron Super v3
Text
Undisclosed
[ChemData 700K]
[nvidia/nvidia/nemotron-3-super-v3]
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
Text
Undisclosed
[In-house data]
[GPT OSS 120B - Apache 2.0]
Synthetic Tool Call Schema for RL
Text
Undisclosed
[In-house data]
[GPT OSS 120B - Apache 2.0]
Synthetic Freeform Text Formatting from GPT-OSS-120B
Text
Undisclosed
[In-house data]
[GPT OSS 120B - Apache 2.0]
Synthetic Citation Formatting from GPT-OSS-120B
Text
Undisclosed
[In-house data]
[GPT OSS 120B - Apache 2.0]
Droid Harness Pivot Vendor Data
Text
Undisclosed
[Droid Harness Pivot vendor data]
Undisclosed
Synthetic HotpotQA Training Data from Qwen3-235B
Text
Undisclosed
[HotpotQA]
[Qwen3-235B]
Synthetic Natural Language Math Proofs from Nemotron 5.5
Text
Undisclosed
[AMC8, AMC10, and AIME problem sets hosted on Art of Problem Solving]; [Pile-StackExchange]
[Nemotron 5.5]
Synthetic Stack Overflow OpenQ
Text
Undisclosed
[Pile-FreeLaw]
Undisclosed
Chemistry Ether0 Vendor Data
Text
Undisclosed
[Chemistry ether0 vendor data]
Undisclosed
Synthetic Litmus-Bench Chemistry from ChEMBL
Text
Undisclosed
[ChEMBL]; [Nemo Gym RL dataset generated from ChEMBL with RDKit]
Undisclosed
Synthetic ZINC Chemistry from Nemotron Super v3
Text
Undisclosed
[ZINC]
[Nemotron Super v3]
ARC-AGI Gym Environment
Text
Undisclosed
[ARC-AGI-2]
[ARC-AGI-2]
Synthetic Agentic Search Tool-Use from DeepSeek-V3.2
Text
Undisclosed
[Mercor Data]
[DeepSeek-V3.2]
Synthetic Text-To-SQL
Text
Undisclosed
[In-house Text-to-SQL data]
[gpt-oss-120b]
Dialog Memory Vendor Data
Text
Undisclosed
[Patronus external vendor agreement]
Undisclosed
Synthetic Indirect Prompt Injection from Nemotron Super v3 and Qwen3-Next-80B-A3B-Instruct
Synthetic Verilog and SystemVerilog Code from DeepSeek-R1-0528 and GPT-OSS-120B
Text
1,233,247
[Verilog/SystemVerilog seed code]
[SDR: DeepSeek R1 0528 and GPT-OSS-120B]; [Filtering: Claude 4 Sonnet]
Synthetic Aider Python Tasks from DeepSeek-R1-0528
Text
236,099
[Exercism (GitHub Python)]
[Deepseek R1 0528]
Synthetic Chat Reasoning-Off Data from GLM-5
Text
646,738
[lmarena-ai/repochat-arena-preference-4k user prompts]
[Multi-turn conversations generated by GLM-5 with best-of-4 selection via Qwen3-Nemotron-235B-A22B-GenRM:]
Synthetic Chat Reasoning-On Data from GLM-5
Text
644,286
[lmarena-ai/repochat-arena-preference-4k user prompts]; [lmarena-ai/arena-expert-5k user prompts]; [lmarena-ai/arena-human-preference-55k user prompts]; [lmarena-ai/arena-human-preference-100k user prompts]; [lmarena-ai/arena-human-preference-140k user prompts]
[Multi-turn conversations generated by GLM-5 with best-of-4 selection via Qwen3-Nemotron-235B-A22B-GenRM:]
Synthetic Multilingual Safety from Riva-Translate-4B-Instruct-v1.1
Text
132,067
[Safety SFT Data: Ultra]
[nvidia/Riva-Translate-4B-Instruct-v1.1]
Synthetic Science Reasoning Effort Medium
Text
502,722
[science-reasoning-effort-medium-v0]
Undisclosed
Synthetic Telecom Tool-Use Trajectories from gpt-oss-120b
Text
12,455
[Existing Tau2 telecom trajectories originally generated with DeepSeek V3.2]
[gpt-oss-120b]
Synthetic Terminal Bench Data from OpenReasoningv2
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 Science MCQ and QA Diversity from GPT-OSS and Kimi-K2
Text
30,358
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Science HLE with Python from GPT-OSS and Kimi-K2
Text
85,184
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Science Search and Python from GPT-OSS and Kimi-K2
Text
6,179
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Science Search from GPT-OSS and Kimi-K2
Text
32,554
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Finance Reasoning from GPT-OSS-120B and Qwen3-235B-A22B-Instruct-2507
Text
326,700
[_SEC filings]
[GPT-OSS-120B, Qwen3-235B-A22B-Instruct-2507]
Synthetic Science Diversity MCQ from GPT-OSS and Kimi-K2
Text
532,942
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Science Diversity OpenQ from GPT-OSS and Kimi-K2
Text
131,045
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Science Reasoning No-Tool from GPT-OSS and Kimi-K2
Text
2,085,600
[doubtnut]; [Pile-FreeLaw]; [Llama Nemotron Dataset]; [askfilo]; [EssentialAI/essential-web-v1.0]; [Vedantu]; [auxiliary_train]; [cdquestions.com]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [AAPT]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]
[GPT-OSS]; [Kimi-K2]
Synthetic Long Context from Qwen3-235B-A22B-Instruct-2507
Text
62,333
[Long-context SFT data: lc_nothink 256k]
[Qwen/Qwen3-235B-A22B-Thinking-2507 and deepseek-ai/DeepSeek-R1]
Synthetic Long Context from Qwen3-235B-A22B-Instruct-2507
Text
49,698
[Long-context SFT data: MRCR 200k]
[Qwen/Qwen3-235B-A22B-Thinking-2507 and deepseek-ai/DeepSeek-R1]
Synthetic Text-To-SQL
Text
96,564
[Undisclosed - no seed data listed]
[gpt-oss-120b]
Synthetic Long Context from Qwen3-235B-A22B-Instruct-2507
Text
397,538
[Long-context SFT data: RULER 256k]
[Qwen/Qwen3-235B-A22B-Thinking-2507 and deepseek-ai/DeepSeek-R1]
For our post-training recipe, we focused on the following languages in addition to English: French, Spanish, Italian, German, Japanese, Hindi, Korean, Brazilian Portuguese, and Chinese
Those languages were represented in the form of multilingual reasoning and translation tasks.
The following table depicts our sample distribution.
Language
Size
English
8.6M
Italian
138k
German
138k
Spanish
138k
French
138k
Japanese
138k
Chinese
138k
Hindi
138k
Korean
138k
Brazilian Portuguese
138k
Evaluation Datasets:
Data Collection Method by dataset
Hybrid: Automated, Human, Synthetic
Labeling Method by dataset
Hybrid: Automated, Human, Synthetic
Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI as outlined in the benchmark section of the model card.
Testing Datasets:
Data Collection Method by dataset
Hybrid: Automated, Human, Synthetic
Labeling Method by dataset
Hybrid: Automated, Human, Synthetic
Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI as outlined in the benchmark section of the model card.
Inference
Acceleration Engine: PyTorch
Test Hardware:
NVIDIA Hopper
H100
H200
NVIDIA Grace Blackwell
GB200
GB300
NVIDIA Blackwell
B200
B300
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_ultra_2026,
2 title = {Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning},
3 author = {{NVIDIA}},
4 year = {2026},
5 url = {https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf},
6 note = {White Paper}
7}
