NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4

Description

NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4 is the quantized release of NVIDIA Nemotron 3.5 Super VL, a vision-language model for reasoning, coding, tool use, document understanding, and video understanding. It uses NVFP4 quantization to reduce memory requirements, with quantization-aware distillation from the BF16 model.

The model combines interleaved Mamba-2, Mixture-of-Experts (MoE), and attention layers with a C-RADIO vision encoder and a vision-language connector. It has 120B total parameters and 12B active parameters and supports context lengths up to 1M tokens. Multi-Token Prediction (MTP) enables native speculative decoding.

This model is ready for commercial or non-commercial use.

Model Summary

Total Parameters	120B (12B active)
Architecture	Mamba-2–Transformer hybrid Mixture of Experts with a C-RADIO vision encoder
Precision	NVFP4 (mixed-precision quantized weights)
Context Length	Up to 1M tokens; deployment settings are listed below
Inputs / Outputs	Text, images, and video / text
Deployment	1× DGX Spark, 1× H200, 2× H100, or 1× GB200
Speculative Decoding	Native Multi-Token Prediction (MTP)
Reasoning Mode	Configurable through the chat template (enable_thinking=True/False)
Recommended Sampling	Text: temperature 1.0, top_p 0.95; vision: temperature 0.6, top_p 0.95
Best For	Efficient deployment of text and vision-language agents
License	OpenMDW-1.1

Model Overview

Model Developer: NVIDIA Corporation

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.

Quick Start

This vLLM configuration is intended for DGX Spark (GB10) only. It serves the model with one concurrent sequence and seven speculative tokens per draft using LiLiCorr.

Nemotron 3.5 Super VL has configurable reasoning levels: off, on (medium), and on. Selecting on (medium) will reduce the number of tokens generated compared to on, but may impact accuracy. You can learn about the reasoning budget here!

Use Docker with NVIDIA GPU support and the vllm/vllm-openai:nightly-3f1bcfd82bb2e93fcd364fc7a5491718453a77e2 or (vllm/vllm-openai:v0.32.0) image. If the model requires authentication, set HF_TOKEN in your shell before launching the container.

export MODEL=nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4
export DRAFT_MODEL=nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4-LiLiCorr

vllm serve $MODEL \
  --served-model-name nemotron-3.5-super-vl \
  --max-num-seqs 1 \
  --trust-remote-code \
  --gpu-memory-utilization 0.75 \
  --max-num-batched-tokens 8192 \
  --speculative_config.model $DRAFT_MODEL\
  --speculative_config.num_speculative_tokens 7 \	
  --speculative_config.method dflash \
  --mamba-backend flashinfer \
  --mamba-ssm-cache-dtype float16 \
  --enable-mamba-cache-stochastic-rounding \
  --mamba-cache-philox-rounds 5 \
  --enable-mamba-shared-prefix-checkpoint \
  --video-pruning-rate 0.5 \
  --media-io-kwargs '{"video": {"fps": 2, "num_frames": 256}}' \
  --block-size 64 \
  --prefix-match-unit 64 \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder \
  --reasoning-parser nemotron_v3

Wait for the server to finish loading the model and compiling kernels, then check the endpoint:

curl --fail http://localhost:8000/v1/models

All vLLM and SGLang examples use port 8000 and the API model name nemotron-3.5-super-vl. See the API Client for text, image, video, and tool-calling examples.

License/Terms of Use

Governing Download Terms: Use of this model is governed by the OpenMDW License Agreement, version 1.1.

Benchmarks

We evaluated our model on the following benchmarks:

Benchmark	Nemotron 3.5 Super VL BF16	Nemotron 3.5 Super VL NVFP4
GPQA-Diamond	89.52	89.90
HLE, text-only	31.79	31.44
MMLU-ProX	76.91	73.76
AA-Omniscience accuracy	26.20	25.86
AA-Omniscience hallucination rate	30.0	27.7
AA-Omniscience index	4.03	5.3
APEX shortlist	88.36	89.36
LiveCodeBench v6	88.22	87.41
SciCode, subtask	47.13	46.61
AA-LCR v1.1	68.62	67.69
GraphWalks	89.71	90.34
IFBench	70.86	71.50
LMArena v3 proxy	52.67	51.93
WMT24++	84.80	84.76
tau2-bench retail	69.24	69.30
tau2-bench airline	76.50	75.75
tau2-bench telecom	73.19	73.19
tau3-bench banking	18.88	17.78
GDPval-AA v2	1211	1206
BrowseComp	78	77.5
SWE-bench Verified	71.10	71.07
SWE-bench Multilingual	67.72	67.00
SWE-bench Pro	51.78	51.30
Terminal-Bench 2.1	60.67	62.78
BabyVision	29.60	29.51
MMLongBench-Doc	62.82	62.82
V*	85.60	85.95
OCR-Reasoning	65.08	65.09
MathVision	78.16	77.79
HLE, full set	29.20	29.00
OCRBench v2, EN	68.12	68.34
OCRBench v2, ZH	59.15	59.03
CharXiv reasoning	80.82	79.88
Video-MME v2	32.85	32.54
OSWorld-Verified	51.14	52.3
BrowseComp [offline]	78	77.5
WebVoyager [offline]	77.72	66.31
AA-Briefcase-Lite [offline]	1137 (-) | 1252 (-) | 40	1137 (-) | 1188 (-) | 45
AutomationBench-AA [offline]	32.44	32.3
GDP.pdf [offline]	6.6 (AA 6.6)	6.0
Terminal Bench 4 [offline]	1	1.5
DeepSWE [offline]	24.65	18.2

The evaluations between the BF16 checkpoint and NVFP4 checkpoint are done with all configs, harnesses, and other non-model parameters held static to ensure fair comparison that isolates the checkpoint impact

We evaluated Nemotron 3.5 Super on a comprehensive suite of benchmarks spanning agentic capabilities, reasoning and knowledge, multimodal, conversational ability and instruction following, long-context understanding, and multilingual performance. All evaluation results for Nemotron 3.5 Super and baselines were collected via NeMo Gym. For reproducibility purposes, more details on the evaluation settings can be found in the NeMo Gym recipes folder and the corresponding reproducibility tutorial. The published recipes specify the benchmark-specific containers, prompts, inference parameters, parser configurations, and scoring settings used to produce the results.

There were a few benchmarks that were not yet onboarded in our open source tools, and for these we used either their official open source implementation or an internal scaffolding that we plan to open source in the future: BabyVision, CharXiv-RQ, MathVision, MMLongBench-Doc, OCR-Reasoning, OCRBench-v2, V*, Video-MME-v2.

Agentic Coding Benchmarks

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

Deployment Geography: Global

Use Case

This model is intended for developers deploying vision-language agents for coding, search, tool-assisted workflows, document question answering, and image and video understanding, on NVIDIA data center and DGX Spark systems.

Release Date

Hugging Face - 10/15/2026 via Hugging Face

Reference(s)

• NVIDIA Nemotron 3.5 Super VL and Lightning Technical Report

Model Architecture

• Architecture Type: Mamba-2–Transformer hybrid Mixture of Experts (MoE) with Multi-Token Prediction (MTP)
• Network Architecture: Nemotron Hybrid MoE with 512 routed experts per MoE layer, a C-RADIO vision encoder, and a vision-language connector
• Number of Model Parameters: 120B total / 12B active

Model Design

The model combines sparse expert computation, attention, and Mamba-2 layers to support long-context processing. The vision encoder and connector turn visual inputs into representations processed by the language backbone. A shared MTP head predicts future tokens recursively for speculative decoding.

The model supports up to 1M tokens of context. The prompt, visual inputs, conversation history, tool results, and generated response share this context window. Individual deployment recipes may use a smaller context window to accommodate the available GPU memory and concurrency.

Training Methodology

Stage 1: Foundation and Multi-Token Prediction

The language backbone is initialized from Nemotron 3 Super Base. A shared MTP head is trained for 100B tokens while the backbone remains frozen, followed by 1T tokens of distillation from Nemotron 3 Ultra Base.

Stage 2: Continued Pre-Training and Mid-Training

The model undergoes 100B tokens of Indic continued pre-training, a 35B-token extension to 1M context, 67B tokens of software-engineering mid-training at 256K context, and a final 9B-token long-context recovery stage. Software-engineering data includes repositories, issue–pull-request pairs, commit diffs, build and CI logs, and agent trajectories.

Stage 3: Vision-Language Training

A C-RADIO vision encoder and connector extend the language backbone to visual inputs, including images, document pages, and video frames.

Stage 4: Supervised Fine-Tuning

A two-stage supervised fine-tuning pipeline jointly trains the language and vision components on vision, reasoning, code, tools, chat, safety, and long-context data. Training uses packed sequences of up to 524,288 tokens.

Stage 5: Reinforcement Learning and Distillation

Reinforcement learning with verifiable rewards, reinforcement learning from human feedback, and specialized teacher training improve reasoning, agentic, chat, and vision capabilities. Multi-teacher on-policy distillation consolidates these capabilities into a single model.

Stage 6: MTP Boosting

The MTP head is further trained against the frozen post-trained backbone to improve speculative-decoding acceptance.

Stage 7: Quantization and Quantization-Aware Distillation

The model is quantized with NVIDIA Model Optimizer and refined through quantization-aware distillation using the supervised fine-tuning data blend. The vision encoder, connector, Mamba convolutions, and embeddings remain in BF16, and the MoE router remains in FP32. Inference uses W4A4 on Blackwell and W4A16 on non-Blackwell platforms, according to the serving configuration.

Input

• Input Type(s): Text, images, and video
• Input Format(s): Text strings, images, and sampled video frames
  • Video: mp4, up to 2 minutes.
  • Image: Red, Green, Blue (RGB) (jpeg, png)
• Input Parameters:
  • Video: Three-Dimensional (3D)
  • Image: Two-Dimensional (2D)
  • Text: One-Dimensional (1D)
• Other Properties Related to Input: Maximum context length up to 1M tokens. Documents can be supplied as page images with accompanying text.

Output

• Output Type(s): Text
• Output Format: String
• Output Parameters: One-dimensional sequences
• Other Properties Related to Output: Input and output tokens share the configured context window.

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.

Software Integration

• Runtime Engine(s): vLLM; SGLang; TensorRT-LLM
• Hardware:
  • NVIDIA Blackwell
  • NVIDIA Hopper
• 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)

• v1.0 - NVFP4 GA

Quick Start Guide

Choose your deployment

Hardware	Runtime	Configuration
1× DGX Spark	vLLM	Text, image; 1M-token context; MTP with 5 draft tokens
2× H100	vLLM	Text, image, and video; 1M-token context; MTP with 3 draft tokens
1× H200	vLLM	Text, image, and video; 1M-token context; MTP with 3 draft tokens
1× GB200	vLLM	Text, image, and video; 1M-token context; MTP with 5 draft tokens
1× GB200, 2× H100 or 1× DGX Spark	SGLang	Text, image; 1M-token context; MTP with 5 draft steps
1× GB200	TensorRT-LLM	1M-token context; MTP with 5 draft tokens

NIM and Dynamo

For production workloads, leverage our best-in-class deployment solutions:

NVIDIA NIM packages the model with an optimized inference engine, validated configurations, and standard APIs for deployment in cloud and data center environments. Get started with the Nemotron 3.5 Super VL NIM here.

NVIDIA Dynamo is a distributed inference framework. It supports KV-cache-aware routing, KV cache management, and disaggregated serving. Get started with Nemotron 3.5 Super VL Dynamo recipes optimized for 7 HW SKUs across Hopper, Blackwell, and Rubin here.

Running on Local Hardware

For Local inference on DGX Spark, RTX Spark, and RTX Pro 6000 you can find details on running the model through the following libraries:

• llama.cpp available here
• Unsloth available here

Speculative Decoding

The serving examples enable the model's native MTP head. No separate drafter download is required for these vLLM and SGLang recipes leveraging MTP. In vLLM, num_speculative_tokens controls draft length; in SGLang, the MTP draft length uses --speculative-num-steps to config.

For DGX Spark, we also provide a LiLiCorr speculative drafter, which accelerates decoding by drafting multiple future tokens in parallel and using a lightweight correlation head to select compatible candidates across adjacent positions. Nemotron Super 3.5 VL verifies the proposed block, allowing generation to advance by multiple accepted tokens per verification step while preserving the target model's output distribution with correct speculative verification. It is recommended to use LiLiCorr for DGX Spark deployments in all scenarios.

Performance depends on hardware, request length, and concurrency. To compare with standard decoding, remove the speculative configuration while keeping the model, workload, sampling parameters, and other server settings the same.

vLLM

Use vllm/vllm-openai:nightly-3f1bcfd82bb2e93fcd364fc7a5491718453a77e2 for these commands (or vllm/vllm-openai:v0.32.0).

2× H100 (up to 1M Context Length)

Use this command if you need to serve the model with up to 1M context length.

vllm serve $MODEL \
  --served-model-name nemotron-3.5-super-vl \
  --tensor-parallel-size 2 \
  --max-num-seqs 128 \
  --trust-remote-code \
  --gpu-memory-utilization 0.89 \
  --max-num-batched-tokens 16384 \
  --block-size 64 \
  --prefix-match-unit 64 \
  --compilation_config.max_cudagraph_capture_size 256 \
  --mamba-backend flashinfer \
  --mamba-ssm-cache-dtype float16 \
  --enable-mamba-cache-stochastic-rounding \
  --mamba-cache-philox-rounds 5 \
  --enable-mamba-shared-prefix-checkpoint \
  --mamba-cache-mode align \
  --speculative-config.method mtp \
  --speculative-config.num_speculative_tokens 3 \
  --video-pruning-rate 0.5 \
  --media-io-kwargs '{"video": {"fps": 2, "num_frames": 256}}' \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder \
  --reasoning-parser nemotron_v3

For one H200, use the same command with --tensor-parallel-size 1.

1× GB200 (up to 1M Context Length)

Use this command if you need to serve the model with up to 1M context length.

vllm serve $MODEL \
  --served-model-name nemotron-3.5-super-vl \
  --trust-remote-code \
  --max-num-seqs 512 \
  --enable-prefix-caching \
  --max-num-batched-tokens 32768 \
  --prefix-match-unit 16 \
  --enable-mamba-shared-prefix-checkpoint \
  --async-scheduling \
  --kernel-config '{"linear_backend_per_quant":{"fp8_w8a8":"torch"}}' \
  --compilation-config '{"pass_config":{"fuse_attn_quant":true},"use_inductor_graph_partition":true,"cudagraph_mode":"FULL_AND_PIECEWISE"}' \
  --mamba-backend flashinfer \
  --mamba-ssm-cache-dtype float16 \
  --enable-mamba-cache-stochastic-rounding \
  --mamba-cache-philox-rounds 5 \
  --mamba-cache-mode align \
  --video-pruning-rate 0.5 \
  --media-io-kwargs '{"video": {"fps": 2, "num_frames": 256}}' \
  --enable-auto-tool-choice \
  --speculative_config.method mtp \
  --speculative_config.num_speculative_tokens 5 \
  --tool-call-parser qwen3_coder \
  --reasoning-parser nemotron_v3

For the multimodal examples, each request can contain one image and one video. The video configuration samples up to 256 frames at 2 FPS with a pruning rate of 0.5. For 1080p videos, use 1 FPS and up to 128 frames; 2 FPS and 256 frames can be used for lower-resolution video such as 720p. Start with short MP4 clips of one to two minutes.

SGLang

Use Docker with NVIDIA GPU support and the lmsysorg/sglang:nemotron3-5-super image.

export MODEL=nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4
export DRAFT_MODEL=nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4-LiLiCorr

1× GB200

sglang serve \
  --model-path $MODEL \
  --host 0.0.0.0 \
  --port 8000 \
  --trust-remote-code \
  --kv-cache-dtype fp8_e4m3 \
  --tp-size 1 \
  --max-running-requests 512 \
  --attention-backend trtllm_mha \
  --mamba-backend flashinfer \
  --mamba-ssm-dtype bfloat16 \
  --enable-mamba-cache-stochastic-rounding \
  --mamba-cache-philox-rounds 5 \
  --mamba-radix-cache-strategy extra_buffer \
  --reasoning-parser nemotron_3 \
  --tool-call-parser qwen3_coder \
  --speculative-algorithm EAGLE \
  --speculative-num-steps 5 \
  --speculative-eagle-topk 1 \
  --speculative-num-draft-tokens 6

2× H100

sglang serve \
  --model-path $MODEL \
  --host 0.0.0.0 \
  --port 8000 \
  --trust-remote-code \
  --kv-cache-dtype fp8_e4m3 \
  --tp-size 2 \
  --max-running-requests 512 \
  --attention-backend flashinfer \
  --mamba-backend flashinfer \
  --mamba-ssm-dtype bfloat16 \
  --enable-mamba-cache-stochastic-rounding \
  --mamba-cache-philox-rounds 5 \
  --mamba-radix-cache-strategy extra_buffer \
  --reasoning-parser nemotron_3 \
  --tool-call-parser qwen3_coder \
  --speculative-algorithm EAGLE \
  --speculative-num-steps 5 \
  --speculative-eagle-topk 1 \
  --speculative-num-draft-tokens 6

1× DGX Spark(use LiLiCorr speculative drafter)

sglang serve \
  --model-path $MODEL \
  --host 0.0.0.0 \
  --port 8000 \
  --trust-remote-code \
  --kv-cache-dtype fp8_e4m3 \
  --tp-size 1 \
  --max-running-requests 1 \
  --attention-backend flashinfer \
  --mamba-backend flashinfer \
  --mamba-ssm-dtype bfloat16 \
  --enable-mamba-cache-stochastic-rounding \
  --mamba-cache-philox-rounds 5 \
  --mamba-radix-cache-strategy extra_buffer \
  --reasoning-parser nemotron_3 \
  --tool-call-parser qwen3_coder \
  --speculative-algorithm DFLASH \
  --speculative-draft-model-path $DRAFT_MODEL

TensorRT-LLM

The following configurations use TensorRT-LLM 1.3.0rc28, built from commit 55f7eca861613bd60e1ac34d5916f091cd89d558 with Python 3.12 and the Mamba cache patch.

Download the NVFP4 checkpoint locally:

hf download "$MODEL_CKPT" --local-dir ./models/super-vl-nvfp4

1× GB200

Save the following as super-vl-gb200.yaml:

tensor_parallel_size: 1
moe_expert_parallel_size: 1
trust_remote_code: true
enable_chunked_prefill: true
disable_overlap_scheduler: false
max_batch_size: 16
max_num_tokens: 8192
max_seq_len: 1048576
attn_backend: TRTLLM
moe_config:
  backend: TRTLLM
kv_cache_config:
  dtype: fp8
  tokens_per_block: 32
  mamba_ssm_cache_dtype: float16
  mamba_ssm_stochastic_rounding: false
  free_gpu_memory_fraction: 0.85
  enable_block_reuse: true
  mamba_state_config:
    periodic_snapshot_interval: 0
    additional_snapshot_offsets_from_start: [50016]
    additional_snapshot_offsets_from_end: [0]
    enable_branch_snapshot: true
cuda_graph_config:
  enable_padding: true
  batch_sizes: [1, 2, 4, 8, 12, 16]
speculative_config:
  decoding_type: MTP
  max_draft_len: 5
guided_decoding_backend: xgrammar

trtllm-serve ./models/super-vl-nvfp4 \
  --config super-vl-gb200.yaml \
  --served_model_name nemotron-3.5-super-vl \
  --host 0.0.0.0 \
  --port 8000 \
  --tool_parser qwen3_coder \
  --reasoning_parser nemotron-v3

max_seq_len sets the context limit; max_num_tokens controls the token batch size. For TensorRT-LLM clients, use the model identifier returned by /v1/models.

1× DGX Spark

Save the following as super-vl-gb200.yaml:

trust_remote_code: true
dtype: bfloat16
tensor_parallel_size: 1
moe_expert_parallel_size: 1
max_batch_size: 2
max_num_tokens: 4096
max_seq_len: 131072

enable_chunked_prefill: true
disable_overlap_scheduler: false
cuda_graph_config: null
attn_backend: TRTLLM

kv_cache_config:
  dtype: fp8
  free_gpu_memory_fraction: 0.9
  max_gpu_total_bytes: 5368709120
  host_cache_size: 0
  mamba_ssm_cache_dtype: float16
  mamba_ssm_stochastic_rounding: false
  enable_block_reuse: true
  use_kv_cache_manager_v2: true
  mamba_state_config:
    enable_branch_snapshot: true

nvfp4_gemm_config:
  allowed_backends: [cutlass, cublaslt, cuda_core]

moe_config:
  backend: CUTEDSL

speculative_config:
  decoding_type: DFlash
  speculative_model: ./models/super-vl-lilicorr
  max_draft_len: 7
  attention_backend: TRTLLM
  context_recompute_tail: 0

guided_decoding_backend: xgrammar

trtllm-serve ./models/super-vl-nvfp4 \
  --config super-vl-spark-lilicorr.yaml \
  --served_model_name nemotron-3.5-super-vl \
  --host 0.0.0.0 \
  --port 8000 \
  --reasoning_parser nemotron-v3

API Client

Install the OpenAI-compatible client:

pip install openai

Text and reasoning

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
model = "nemotron-3.5-super-vl"

response = client.chat.completions.create(
    model=model,
    messages=[{"role": "user", "content": "Explain why binary search takes logarithmic time."}],
    temperature=1.0,
    top_p=0.95,
    max_tokens=16384,
    extra_body={"chat_template_kwargs": {"enable_thinking": True}},
)
choice = response.choices[0]
if choice.finish_reason == "length":
    raise RuntimeError("The response reached max_tokens; increase the output budget.")
print(choice.message.content)

Set enable_thinking to False to disable reasoning. For medium reasoning effort, use {"enable_thinking": True, "medium_effort": True}. Reasoning tokens and the final answer share the max_tokens budget.

Images

Use a multimodal server configuration. The example below sends a local image as a base64 data URL.

import base64
from pathlib import Path
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
image_bytes = Path("image.png").read_bytes()
image_url = "data:image/png;base64," + base64.b64encode(image_bytes).decode()

response = client.chat.completions.create(
    model="nemotron-3.5-super-vl",
    messages=[{"role": "user", "content": [
        {"type": "text", "text": "Describe the image and any visible text."},
        {"type": "image_url", "image_url": {"url": image_url}},
    ]}],
    temperature=1.0,
    top_p=0.95,
    max_tokens=16384,
)
if response.choices[0].finish_reason == "length":
    raise RuntimeError("The response reached max_tokens; increase the output budget.")
print(response.choices[0].message.content)

Video with vLLM

Use a multimodal vLLM configuration with the video sampling settings above.

import base64
from pathlib import Path
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
video_bytes = Path("video.mp4").read_bytes()
video_url = "data:video/mp4;base64," + base64.b64encode(video_bytes).decode()

response = client.chat.completions.create(
    model="nemotron-3.5-super-vl",
    messages=[{"role": "user", "content": [
        {"type": "text", "text": "Summarize the events in this video in order."},
        {"type": "video_url", "video_url": {"url": video_url}},
    ]}],
    temperature=1.0,
    top_p=0.95,
    max_tokens=16384,
)
if response.choices[0].finish_reason == "length":
    raise RuntimeError("The response reached max_tokens; increase the output budget.")
print(response.choices[0].message.content)

Tool calling

The model requests a tool call; the application executes it and returns the result. This complete example uses a local inventory lookup.

import json
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
model = "nemotron-3.5-super-vl"
inventory = {"keyboard": 12, "monitor": 5}
tools = [{"type": "function", "function": {
    "name": "get_inventory",
    "description": "Return the inventory count for a product.",
    "parameters": {
        "type": "object",
        "properties": {"product": {"type": "string", "enum": ["keyboard", "monitor"]}},
        "required": ["product"],
        "additionalProperties": False,
    },
}}]
messages = [{"role": "user", "content": "How many keyboards are in stock?"}]

def complete(history, tool_choice="auto"):
    response = client.chat.completions.create(
        model=model,
        messages=history,
        tools=tools,
        tool_choice=tool_choice,
        temperature=1.0,
        top_p=0.95,
        max_tokens=16384,
        extra_body={"chat_template_kwargs": {"enable_thinking": True}},
    )
    choice = response.choices[0]
    if choice.finish_reason == "length":
        raise RuntimeError("The response was truncated; increase max_tokens.")
    return choice.message

message = complete(messages)
messages.append(message.model_dump(exclude_none=True))
if not message.tool_calls:
    print(message.content)
else:
    for call in message.tool_calls:
        arguments = json.loads(call.function.arguments)
        if (call.function.name != "get_inventory"
                or not isinstance(arguments, dict)
                or set(arguments) != {"product"}
                or not isinstance(arguments["product"], str)
                or arguments["product"] not in inventory):
            raise ValueError("Invalid inventory tool call")
        product = arguments["product"]
        messages.append({
            "role": "tool",
            "tool_call_id": call.id,
            "content": json.dumps({"product": product, "count": inventory[product]}),
        })
    print(complete(messages, tool_choice="none").content)

Training, Testing, and Evaluation Datasets

Training

Data Modality: Text, images, and video
Foundation Training Data Size: More than 20 trillion tokens Image Training Data Size: 1 Million to 1 Billion Images Text Training Data Size: More than 10 Trillion Tokens Video Training Data Size: 10,000 to 1 Million Hours Foundation Data Collection Period: 2013 to December 2025
Data Collection Method: Hybrid: automated, manually-collected, and synthetic
Labeling Method: Hybrid: automated, manually-labeled, and synthetic

The foundation corpus contains curated and synthetic text covering English, 19 other spoken languages, and 43 programming languages. Sources include webpages, dialogue, articles, code, and other written materials across math, science, legal, finance, and general knowledge domains.

The post-training corpus includes reasoning, coding, tool use, instruction following, structured outputs, chat, safety, and long-context tasks, together with image and video understanding. Synthetic trajectories are filtered for structural correctness, repetition, and task validity.

For more information about the datasets used to train this model, please see the Public Summary of Training Content

Detailed dataset information

Base Pre-Training Corpus (Nemotron 3 Foundation)

The foundation of the model is trained on the Nemotron 3 corpus, comprising the following datasets from the Nemotron Pretraining Datasets collection:

Dataset Collection	Token Counts	Description
Nemotron-CC-v2 & v2.1	9.1T	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.
Nemotron-Pretraining-Specialized-v1 & v1.1 & v1.2 & Nemotron-Pretraining-SFT-v1	660.0B	Synthetic datasets targeting specialized domains such as STEM reasoning and scientific coding.
Nemotron-Pretraining-Legal-v1	4.3B	Synthetic datasets targeting the legal domain.

Public Datasets

Dataset	Collection Period
GSM8K	4/23/2025
CC-NEWS	4/23/2025
Common Crawl	4/23/2025
Wikimedia	4/23/2025
Bespoke-Stratos-17k	4/23/2025
tigerbot-kaggle-leetcodesolutions-en-2k	4/23/2025
glaive-function-calling-v2	4/23/2025
APIGen Function-Calling	4/23/2025
LMSYS-Chat-1M	4/23/2025
Open Textbook Library - CC BY-SA & GNU subset and OpenStax - CC BY-SA subset	4/23/2025
Advanced Reasoning Benchmark, tigerbot-kaggle-leetcodesolutions-en-2k, PRM800K, and SciBench	4/23/2025
FineWeb-2	4/23/2025
Court Listener	Legacy Download
peS2o	Legacy Download
OpenWebMath	Legacy Download
BioRxiv	Legacy Download
PMC Open Access Subset	Legacy Download
OpenWebText2	Legacy Download
Stack Exchange Data Dump	Legacy Download
PubMed Abstracts	Legacy Download
NIH ExPorter	Legacy Download
arXiv	Legacy Download
BigScience Workshop Datasets	Legacy Download
SEC's Electronic Data Gathering, Analysis, and Retrieval (EDGAR)	Legacy Download
Advanced Mathematical Problem Solving	Legacy Download
MathPile	Legacy Download
NuminaMath CoT	Legacy Download
PMC Article	Legacy Download
FLAN	Legacy Download
Advanced Reasoning Benchmark	Legacy Download
SciBench	Legacy Download
WikiTableQuestions	Legacy Download
FinQA	Legacy Download
Riddles	Legacy Download
Problems in Elementary Mathematics for Home Study	Legacy Download
MedMCQA	Legacy Download
Cosmos QA	Legacy Download
MCTest	Legacy Download
AI2's Reasoning Challenge	Legacy Download
OpenBookQA	Legacy Download
MMLU Auxiliary Train	Legacy Download
social-chemestry-101	Legacy Download
Moral Stories	Legacy Download
The Common Pile v0.1	Legacy Download
FineMath	Legacy Download
MegaMath	Legacy Download
MultiverseMathHard	10/2/2025
SWE-Gym	10/2/2025
WorkBench	10/2/2025
WildChat-1M	10/2/2025
OpenCodeReasoning-2	10/2/2025
HelpSteer3	10/2/2025
opc-sft-stage2	10/2/2025
Big-Math-RL-Verified	10/2/2025
MetaMathQA	10/2/2025
simple-arithmetic-problems	10/2/2025
arithmetic	10/2/2025
Skywork-OR1-RL-Data	10/2/2025
FastChat	10/2/2025
News Commentary	10/2/2025
Essential-Web	10/2/2025
finepdfs	10/2/2025
HotpotQA	10/2/2025
SQuAD2.0	10/2/2025
NLTK Words Lists	10/2/2025
NVClip (NV-CC-Img-Text-Dataset)	2022-2026
ScaleCUA, Processed Training Subset	2022-2026
RoboInter-VQA	2022-2026
Open Images V6, Object-Detection Training Records	2022-2026
YT1B PLM Video QA, Custom No-Thinking and Safety-Filtered	2022-2026
YT1B PLM Dense Captions, No-Thinking and Safety-Filtered	2022-2026
RoboVQA	2022-2026
YT1B PLM Thinking Traces, Safety-Filtered	2022-2026
Boyi Short-Video QA	2022-2026
Boyi Short-Video QA, Custom No-Thinking and Safety-Filtered	2022-2026
K710 Video Classification, Full Training Snapshot	2022-2026
Molmo2-VideoCapQA	2022-2026
Molmo2-VideoCapQA, Custom No-Thinking and Safety-Filtered	2022-2026
Molmo2-VideoPoint, Custom No-Thinking and Safety-Filtered	2022-2026
Molmo2-VideoPoint	2022-2026

Crawled and Scraped from Online Sources by NVIDIA

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 used in the Nemotron-CC data pipeline. 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.

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
Sangraha	Text	109,263,115 indexed source sequences: 77,231,480 verified and 32,031,635 unverified	Undisclosed	NVIDIA Advanced Deep Learning Research
LITA Chunked Video Labels, Custom No-Thinking Dense Captions	Text (video labels)	151,349 records	Undisclosed	NVIDIA Advanced Deep Learning Research
LITA Chunked Video Labels, Cleaned v2	Text (video labels)	150,434 records	Undisclosed	NVIDIA Advanced Deep Learning Research
Single-Page Document Labels, Qwen3.5 v2	Text and images	50,930 records	Undisclosed	NVIDIA Advanced Deep Learning Research
Processed YouTube Videos, Cosmos Dense Captions	Text and video	274,472 records	Undisclosed	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
Mercor SWE-AgentsV1	Undisclosed
KeenSight Batch 1, Relabeled Subset	Qwen3.5-122B-A10B
KeenSight Batch 1, Remaining Subset	Qwen3.5-122B-A10B
Camio Multilingual OCR and Document Layout	Undisclosed
Chinese Visual Question Answering, Relabeled Subset	Qwen3.5-122B-A10B
Camio Chinese OCR, Cleaned Variant	Qwen3.5-122B-A10B
Camio English OCR, Cleaned Variant	Qwen3.5-122B-A10B
KeenSight Batch 2, Remaining Subset	Qwen3.5-122B-A10B
KeenSight Batch 1, Safety-Filtered Trace Variant	Qwen3.5-122B-A10B
Chinese Visual Question Answering, Safety-Filtered Trace Variant	Qwen3.5-122B-A10B
Camio Chinese Question Answering, Verified-Correct Subset	Qwen3.5
KeenSight Chinese OCR, Filtered Subset	DeepSeek-OCR
Camio Synthetic Trading-Graph Question Answering	Undisclosed
Camio Synthetic Menu and Receipt Question Answering	Qwen3.5-122B-A10B
Camio Chinese OCR, Formatted Variant	Qwen3.5-122B-A10B
Camio Synthetic Time-Series Question Answering	Undisclosed
Camio English OCR, Formatted Variant	Qwen3.5-122B-A10B
Lean Formal Proofs — RL	Undisclosed
Document-Grounded Finance Reasoning, Patch Tasks — RL	Undisclosed
Math CoT — SFT	Kimi K3
Chat — SFT	Undisclosed
Chat, Reasoning Off — SFT	Undisclosed
Conversational Tool-Use Simulation — RL	Undisclosed
Math CoT, Low Variant — SFT	Kimi K3
Document-Grounded Finance Reasoning — SFT	Undisclosed
Safety Preference v0.3 — RL	Undisclosed
Response-Length Instruction Following, Reasoning Off — SFT	Undisclosed
Math Python TIR, Low Variant — SFT	Kimi K3
Math Python TIR — SFT	Kimi K3
General-Purpose Tool Calling — SFT	Qwen3.5-397B-A17B; Gemma-4-31B-it
Knowledge Agent — SFT	Undisclosed
DataFunnel arXiv Science4 v1.0 — RL	DeepSeek-V4-Flash (problem extraction); GLM-5.2 (refinement); Kimi K3 and K3 V5 (verification and filtering); Super 3.5 r14 (difficulty profiling); Qwen3-235B-A22B-Instruct-2507-FP8 (reward judging)
Structured Outputs v3 — RL	gpt-oss-120b
Response-Length Instruction Following — SFT	Undisclosed
Filtered CUDA Coding — SFT	Undisclosed
Reasoning Gym — RL	Undisclosed
Competitive Coding Challenges v2 — SFT	GLM-5.2
Multi-Turn Instruction Following, Reasoning Off — SFT	Undisclosed
Multi-Turn Instruction Following — SFT	Undisclosed
Instruction Following, Filtered RLHF Prompts — SFT	Undisclosed
Citation Formatting — RL	gpt-oss-120b
Long-Context General Data — SFT	Undisclosed
Free-Form Formatting — RL	gpt-oss-120b
Structured Outputs v4 — RL	Undisclosed
Tool-Call Schema — RL	No model used for RL conversion; source-generation models undisclosed
Competitive Coding, OCR Python — SFT	DeepSeek-V4-Pro
Lean Formal Proofs — SFT	gpt-oss-120b; Goedel-Prover-V2-32B; Qwen2.5-32B-Instruct
Citation Instruction Following — RL	Undisclosed
Search Graph Walks — SFT	MiniMax-M2
General-Purpose Tool Calling, Reasoning Off — SFT	Undisclosed
Knowledge Agent Tau3, Reasoning Off — SFT	Undisclosed
ARC-AGI Puzzle Reasoning — SFT	Kimi-K2.5; Qwen3-235B-A22B-Thinking-2507; DeepSeek-V3.2; Qwen3-Next-80B-A3B-Thinking; GLM-4.5-Air-FP8; gpt-oss-120b; GLM-4.7-FP8; Ring-1T; Ministral-3-14B-Reasoning-2512
Calendar — RL	Undisclosed
ACEProof — SFT	Undisclosed
Long Translation — SFT	Undisclosed
Agentless Software Engineering — SFT	Undisclosed
Long-Context PDF Translation v1.0 — SFT	DeepSeek-V4-Flash (translation generation); GPT-5.6 Sol (filtering and judging)
KeenSight Batch 2, Relabeled and Filtered Subset	Qwen3.5-122B-A10B; Gemini 3 Flash Preview (filtering)
Camio Chinese Letter QA, Cleaned entrace_cleaned Variant	Qwen3-VL-30B-A3B-Thinking-FP8; gpt-oss-120b; Qwen3.5-122B-A10B; GPT-4o mini (filtering and quality checks)
Camio Chinese Letter QA, Cleaned Variant	Qwen3-VL-30B-A3B-Thinking-FP8; gpt-oss-120b; Qwen3.5-122B-A10B; GPT-4o mini (filtering and quality checks)
Camio Chinese Text QA, Cleaned entrace_cleaned Variant	Qwen3-VL-30B-A3B-Thinking-FP8; gpt-oss-120b; Qwen3.5-122B-A10B; GPT-4o mini (filtering and quality checks)

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	Undisclosed
Machine Translation of STEM -	Qwen2.5-14B-Instruct
Competitive Coding RL data from Nemotron Cascade	Undisclosed
Long context RL	Undisclosed
Single-step SWE RL for patch generation	Undisclosed
OpenHands SWE	Undisclosed

NVIDIA-Sourced Synthetic Datasets (Pre-Training)

Dataset	Modality	Dataset Size	Seed Dataset	Model(s) used for generation
Nemotron-Pretraining-Fact-Seeking	Text	35.0B	FineWiki	Qwen3-30B-A3B-Instruct-2507
Nemotron-Pretraining-Legal	Text	4.3B	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)	Qwen3-235B-A22B-Thinking-2507
Nemotron-Pretraining-Formal-Logic	Text	128M	Nemotron Personas	Qwen3-235B-A22B-Thinking-2507
Nemotron-Pretraining-Economics	Text	73.4M	-	Qwen3-235B-A22B-Thinking-2507
Nemotron-Pretraining-Multiple-Choice	Text	1.6B	MMLU Auxiliary Train	DeepSeek-V3; Qwen3-235B-A22B
Nemotron-Pretraining-Code-Concepts	Text	7.3B	-	gpt-oss-20b; gpt-oss-120b
Nemotron-Pretraining-Unconditional-Algorithmic	Text	196.5M	-	gpt-oss-120b; Qwen3-235B-A22B
More Synthetic Tasks from DeepSeek-V3 and Qwen3-235B-A22B	Text	1.1B	train splits of acp_bench; ai2_arc; babi; gsm8k; hendrycks_math; IFEval; MedText; mediqa_qa; mlqa; MMLU-Pro; mmlu-pro-plus; MMLU-ProX; nq_open; tinyGSM8k; truthful_qa; truthfulqa-multi; MATH-lighteval; mmlu; awesome-chatgpt-prompts; super_glue	DeepSeek v3; Qwen3-235B-A22B
Synthetic Tasks from DeepSeek-V3 and Qwen3-235B-A22B	Text	6.7B	train splits of Into the Unknown; AI2 ARC (AI2 Reasoning Challenge); BLiMP (Benchmark of Linguistic Minimal Pairs); CommonSenseQA; GLUE; HeadQA; Hendrycks Ethics; Memo Trap; modus-tollens; NeQA; pattern-matching-suppression; mastermind_24_mcq_random; mastermind_24_mcq_close; quote-repetition; redefine-math; Repetitive Algebra; sig-figs; MMLU-Pro; MC-TACO; MedConceptsQA; MMLU_dataset; OpenbooksQA; PIQA (Physical Interaction Question Answering); SocialIQA; SuperGLUE; tinyAI2_arc; tinyMMLU; tinyWinogrande; TruthfulQA; WebQuestions; Winogrande; GPQA; MBPP	DeepSeek v3; Qwen3-235B-A22B
Synthetic Art of Problem Solving from DeepSeek-R1	Text	40B	Art of Problem Solving; American Mathematics Competitions 8; American Mathematics Competitions 10	DeepSeek-R1
Synthetic Moral Stories and Social Chemistry from Qwen3-235B-A22B-Thinking-2507 and Mixtral-8x22B-v0.1	Text	15.2M	social-chemestry-101; Moral Stories	Qwen3-235B-A22B-Thinking-2507; Mixtral-8x22B-v0.1
Synthetic Moral Stories and Social Chemistry from Mixtral-8x22B-v0.1	Text	327M	social-chemestry-101; Moral Stories	Mixtral-8x22B-v0.1
Synthetic Social Sciences seeded with OpenStax from DeepSeek-V3, Mixtral-8x22B-v0.1, and Qwen2.5-72B	Text	83.6M	OpenStax - CC BY-SA subset	DeepSeek-V3; Mixtral-8x22B-v0.1; Qwen2.5-72B
Synthetic Health Sciences seeded with OpenStax from DeepSeek-V3, Mixtral-8x22B-v0.1, and Qwen2.5-72B	Text	9.7M	OpenStax - CC BY-SA subset	DeepSeek-V3; Mixtral-8x22B-v0.1; Qwen2.5-72B
Synthetic STEM seeded with OpenStax, Open Textbook Library, and GSM8K from DeepSeek-R1, DeepSeek-V3, DeepSeek-V3-0324, and Qwen2.5-72B	Text	175M	OpenStax - CC BY-SA subset; GSM8K; Open Textbook Library - CC BY-SA & GNU subset	DeepSeek-R1, DeepSeek-V3; DeepSeek-V3-0324; Qwen2.5-72B
Nemotron-PrismMath	Text	4.6B	Big-Math-RL-Verified; OpenR1-Math-220k	Qwen2.5-0.5B-instruct, Qwen2.5-72B-Instruct; DeepSeek-R1-Distill-Qwen-32B
Synthetic Question Answering Data from Papers and Permissible Books from Qwen2.5-72B-Instruct	Text	350M	arXiv; National Institutes of Health ExPorter; BioRxiv; PMC Article; USPTO Backgrounds; peS2o; Global Regulation; CORE; PG-19; DOAB CC BY & CC BY-SA subset; NDLTD	Qwen2.5-72B-Instruct
Synthetic Rephrased Math Data from Common Crawl from phi-4	Text	73B	Common Crawl	phi-4
Synthetic Math Data from Common Crawl 4plus	Text	52.3B	Common Crawl	phi-4
Synthetic Math Data from Common Crawl 3	Text	80.9B	Common Crawl	phi-4
Synthetic AGIEval seeded with AQUA-RAT, LogiQA, and AR-LSAT from DeepSeek-V3 and DeepSeek-V3-0324	Text	4.0B	AQUA-RAT; LogiQA; AR-LSAT	DeepSeek-V3; DeepSeek-V3-0324
Synthetic AGIEval seeded with AQUA-RAT, LogiQA, and AR-LSAT from Qwen3-30B-A3B	Text	4.2B	AQUA-RAT; LogiQA; AR-LSAT	Qwen3-30B-A3B
Synthetic Art of Problem Solving from Qwen2.5-32B-Instruct, Qwen2.5-Math-72B, Qwen2.5-Math-7B, and Qwen2.5-72B-Instruct	Text	Undisclosed	Art of Problem Solving; American Mathematics Competitions 8; American Mathematics Competitions 10; GSM8K; PRM800K	Qwen2.5-32B-Instruct; Qwen2.5-Math-72B; Qwen2.5-Math-7B; Qwen2.5-72B-Instruct
Synthetic MMLU Auxiliary Train from DeepSeek-R1	Text	0.5B	MMLU Auxiliary Train	DeepSeek-R1
Synthetic Long Context Continued Post-Training Data from Papers and Permissible Books from Qwen2.5-72B-Instruct	Text	Undisclosed	arXiv; National Institutes of Health ExPorter; BioRxiv; PMC Article; USPTO Backgrounds; peS2o; Global Regulation; CORE; PG-19; DOAB CC BY & CC BY-SA subset; NDLTD	Qwen2.5-72B-Instruct
Synthetic Common Crawl from Qwen3-30B-A3B and Mistral-Nemo-12B-Instruct	Text	415.8B	Common Crawl	Qwen3-30B-A3B; Mistral-NeMo-12B-Instruct
Synthetic Multilingual Data from Common Crawl from Qwen3-30B-A3B	Text	Undisclosed	Common Crawl	Qwen3-30B-A3B
Synthetic Multilingual Data from Wikimedia from Qwen3-30B-A3B	Text	Undisclosed	Wikimedia	Qwen3-30B-A3B
Synthetic Math Data from Wikimedia from Nemotron-4-340B-Instruct	Text	Undisclosed	-	Nemotron-4-340B-Instruct
Synthetic Common Crawl Code from phi-4	Text	427.9B	Common Crawl	phi-4
Synthetic Scientific Coding from Qwen3-235B-A22B	Text	1.2B	Wikimedia	Qwen3-235B-A22B
Tool Calling Data	Text	26.2B	-	Qwen3-235B-A22B-2507; gpt-oss-120b
Synthetic Essential-Web from QwQ-32B	Text	28.1B	Essential-Web	QwQ-32B
Translated Synthetic Crawl	Text	389.9B	Common Crawl	Qwen3-30B-A3B
Translated Synthetic Wikipedia	Text	7.9B	Wikimedia	Qwen3-30B-A3B
Synthetic Long Context from Qwen3-235B-A22B-Instruct-2507	Text	Undisclosed	CORE; PG-19; DOAB CC BY & CC BY-SA subset; NDLTD	Qwen3-235B-A22B-Instruct-2507
Synthetic Search STEM OPENQ from DeepSeek-R1-0528	Text	Undisclosed	-	DeepSeek-R1-0528
Synthetic MCQ from Qwen2.5-32B-Instruct and DeepSeek-R1-0528	Text	Undisclosed	-	Qwen2.5-32B-Instruct; DeepSeek-R1-0528
Synthetic Offline Search MCQA HLE from DeepSeek-R1-0528	Text	Undisclosed	-	DeepSeek-R1-0528
Synthetic Offline Search MCQA GPQA from Qwen3-235B-A22B and DeepSeek-R1-0528	Text	Undisclosed	-	Qwen3-235B-A22B; DeepSeek-R1-0528
Synthetic Human Preference from QwQ-32B, Qwen3-30B-A3B, Qwen3-235B-A22B, Qwen3-235B-A22B-Instruct-2507, Mistral-Small-3.1-24B-Instruct-2503, Mistral-Small-3.2-24B-Instruct-2506, MiniMax-M1-80k, MiniMax-M1-40k, Kimi-K2-Instruct, DeepSeek-V3-0324, DeepSeek-R1-0528	Text	Undisclosed	-	QwQ-32B; Qwen3-30B-A3B; Qwen3-235B-A22B; Qwen3-235B-A22B-Instruct-2507; Mistral-Small-3.1-24B-Instruct-2503; Mistral-Small-3.2-24B-Instruct-2506; MiniMax-M1-80k; MiniMax-M1-40k; Kimi-K2-Instruct; DeepSeek-V3-0324; DeepSeek-R1-0528
Synthetic WildChat-1M and arena-human-preference-140k from DeepSeek-R1, gemma-2-2b-it, gemma-3-27b-it, gpt-oss-20b, gpt-oss-120b, Mistral-7B-Instruct-v0.3, Mixtral-8x22B-Instruct-v0.1, Nemotron-4-340B-Instruct, NVIDIA-Nemotron-Nano-9B-v2, Phi-4-mini-instruct, Phi-3-small-8k-instruct, Phi-3-medium-4k-instruct, Qwen3-235B-A22B, QwQ-32B	Text	Undisclosed	WildChat-1M; arena-human-preference-140k	DeepSeek-R1; gemma-2-2b-it; gemma-3-27b-it; gpt-oss-20b; gpt-oss-120b; Mistral-7B-Instruct-v0.3; Mixtral-8x22B-Instruct-v0.1; Nemotron-4-340B-Instruct; NVIDIA-Nemotron-Nano-9B-v2; Phi-4-mini-instruct; Phi-3-small-8k-instruct; Phi-3-medium-4k-instruct; Qwen3-235B-A22B; QwQ-32B
Synthetic Code from Qwen3-32B	Text	Undisclosed	English Common Crawl; English Common Crawl 1.1	Qwen3-32B
Synthetic OpenCodeReasoning from DeepSeek-R1	Text	Undisclosed	OpenCodeReasoning	DeepSeek-R1
Synthetic OpenCodeReasoning from DeepSeek-R1-0528	Text	Undisclosed	OpenCodeReasoning	DeepSeek-R1-0528
Synthetic HackerRank Coding from DeepSeek-R1-0528	Text	Undisclosed	HackerRank Coding Dataset	DeepSeek-R1-0528
Synthetic LIMO from DeepSeek-R1-0528	Text	Undisclosed	LIMO	DeepSeek-R1-0528
Synthetic SCP from DeepSeek-R1-0528	Text	Undisclosed	SCP-116K	DeepSeek-R1-0528
Synthetic Stack Exchange from DeepSeek-R1-0528	Text	Undisclosed	Stack Exchange	DeepSeek-R1-0528
Synthetic Stack Exchange from gpt-oss-120b and Qwen2.5-32B-Instruct	Text	Undisclosed	Stack Exchange	gpt-oss-120b; Qwen2.5-32B-Instruct
Synthetic Stack Exchange from gpt-oss-120b	Text	Undisclosed	Stack Exchange	gpt-oss-120b
Synthetic Art of Problem Solving from gpt-oss-120b and Qwen2.5-32B-Instruct	Text	Undisclosed	Art of Problem Solving; American Mathematics Competitions 8; American Mathematics Competitions 10	gpt-oss-120b; Qwen2.5-32B-Instruct
Synthetic Common Crawl from Qwen3-30B-A3B	Text	Undisclosed	Common Crawl	Qwen3-30B-A3B
Synthetic Wikipedia from Qwen3-30B-A3B	Text	Undisclosed	Wikimedia	Qwen3-30B-A3B
Synthetic Essential-Web from Qwen3-30B-A3B and Qwen3-235B-A22B-Thinking-2507	Text	Undisclosed	Essential-Web	Qwen3-30B-A3B; Qwen3-235B-A22B-Thinking-2507
Synthetic Essential-Web from gpt-oss-120b	Text	Undisclosed	Essential-Web	gpt-oss-120b
Synthetic Textbook Math from Qwen3-30B-A3B, Qwen3-235B-A22B, phi-4	Text	Undisclosed	Common Crawl; FineMath	Qwen3-30B-A3B; Qwen3-235B-A22B; phi-4
Synthetic Math and Code from DeepSeek-R1 and DeepSeek-R1-0528	Text	Undisclosed	Magicoder-Evol-Instruct-110K; opc-sft-stage2; TACO; OpenCodeReasoning; OpenMathReasoning; NuminaMath CoT	DeepSeek-R1; DeepSeek-R1-0528
Synthetic Math from gpt-oss-120b and Qwen2.5-32B-Instruct	Text	Undisclosed	-	gpt-oss-120b; Qwen2.5-32B-Instruct
Synthetic OpenMathReasoning from gpt-oss-120b and Qwen2.5-32B-Instruct	Text	Undisclosed	OpenMathReasoning	gpt-oss-120b; Qwen2.5-32B-Instruct
Synthetic KernelBook from DeepSeek-R1-0528	Text	Undisclosed	KernelBook	DeepSeek-R1-0528
Synthetic Scale HLE from gpt-oss-120b	Text	Undisclosed	Scale HLE	gpt-oss-120b
Synthetic CDQuestions from gpt-oss-120b	Text	Undisclosed	CDQuestions	gpt-oss-120b
Synthetic GPQA from gpt-oss-120b and Qwen2.5-32B-Instruct	Text	Undisclosed	Stack Exchange	gpt-oss-120b; Qwen2.5-32B-Instruct
Synthetic Vedantu from gpt-oss-120b	Text	Undisclosed	Vedantu	gpt-oss-120b
Synthetic Search STEM MCQ from Qwen3-235B-A22B and DeepSeek-R1-0528	Text	Undisclosed	-	Qwen3-235B-A22B; DeepSeek-R1-0528
Synthetic OpenSTEM from Qwen2.5-32B-Instruct and DeepSeek-R1-0528	Text	Undisclosed	-	Qwen2.5-32B-Instruct; DeepSeek-R1-0528
Synthetic MCQ10 from DeepSeek-R1-0528	Text	Undisclosed	-	DeepSeek-R1-0528
Synthetic MCQ4 from Qwen3-235B-A22B, DeepSeek-R1-0528, and Qwen3-235B-A22B-Instruct-2507	Text	Undisclosed	-	Qwen3-235B-A22B; DeepSeek-R1-0528; Qwen3-235B-A22B-Instruct-2507
Indic SDG for Nemotron	Text	711,276,565 indexed source-sequence entries; final training-blend count not confirmed	nvidia/Nemotron-CC-v2	google/gemma-4-31B-it; google/gemma-4-31B-it-assistant

NVIDIA-Sourced Synthetic Datasets (Post-Training)

Dataset	Modality	Dataset Size	Seed Dataset	Model(s) used for generation
Synthetic Single-Step Tool Use — RL	Text	Approximately 1,433,203 samples	Handwritten prompts; no external seed data	DeepSeek-V3.2; DeepSeek-R1-0528; Qwen3-235B-A22B-Thinking-2507; Qwen3-32B
Synthetic Sparse Tables with Mixed Prompts	Text and images	100,000 records	Generated in-house; no external seed data	None; generated without an SDG model
Synthetic Single-Step Tool Use — RL	Text	Approximately 1,433,203 samples	Handwritten prompts; no external seed data	DeepSeek-V3.2; DeepSeek-R1-0528; Qwen3-235B-A22B-Thinking-2507; Qwen3-32B
Synthetic Sparse Tables with Mixed Prompts	Text and images	100,000 records	Generated in-house; no external seed data	None; generated without an SDG model
Synthetic Competitive MATH Proofs from DeepSeek-V4-Pro	Text	Undisclosed	[AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]	[deepseek-ai/DeepSeek-V4-Pro]
Synthetic Hermes Agent Reasoning Traces	Text	Undisclosed	[lambda/hermes-agent-reasoning-traces]	[hermes-agent-generator]
Synthetic Competitive Coding from DeepSeek-V4-Pro	Text	Undisclosed	[NVCompetitiveCodingV1]	[deepseek-ai/DeepSeek-V4-Pro]
Synthetic Competitive Science Reasoning from DeepSeek-V4-Pro	Text	Undisclosed	[AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]; [EssentialAI/essential-web-v1.0]; [cdquestions.com]; [Pile-FreeLaw]; [Vedantu]; [askfilo]; [doubtnut]; [ICHO-IPH0 Dataset]; [LIMO dataset (Less is More for Reasoning)]; [AAPT]; [ChemData 700K]; [oMeBench]; [Flavor Analysis and Recognition Transformer]; [ChemCoTBench]; [Llama Nemotron Dataset]	[deepseek-ai/DeepSeek-V4-Pro]
Synthetic Competitive MATH CoT and TIR from Nemotron 5.5	Text	Undisclosed	[Pile-FreeLaw]; [AMC 8 Problems and Solutions, AMC 10 Problems and Solution, and AIME Problems and Solutions]	[Nemotron 5.5]
Vendor Terminal Bench-like Tasks from Mercor	Text	Undisclosed	[Terminal bench like tasks curated by the vendor]	[Undisclosed - purchased dataset]
Turing Math Data Pack	Text	Undisclosed	[Turing Math Data Pack dataset]	[Undisclosed - purchased dataset]
Synthetic Holdout, Skywork, DAPO, and Turing Math from GPT-5.5	Text	Undisclosed	[DocQA-RL-1.6K]; [DAPO-Math-17k]	[GPT-5.5]
Synthetic Long Context RL from QwenLong L1 and DocQA-RL-1.6K	Text	Undisclosed	[DocQA-RL-1.6K]	Undisclosed
Synthetic Competitive Coding Gym Tasks	Text	Undisclosed	[NVCompetitiveCodingV1.1]	Undisclosed
Synthetic Finance SEC Search Agent from GPT-OSS-120B and Qwen3	Text	Undisclosed	[SEC filings from sec.gov]	[GPT-OSS-120B]; [Qwen3-235B-A22B-Instruct]; [Qwen3-4B-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	Text	Undisclosed	[Nemotron-RL-agent-structured-outputs-v1]	[Qwen3-30B-A3B-Instruct-2507]; [Qwen3-235B-A22B-Instruct-2507]
Synthetic Long Context Equivalence Rule from Qwen3-235B-A22B-Thinking-2507 and DeepSeek-R1	Text	Undisclosed	[Long-context SFT data]	[Qwen/Qwen3-235B-A22B-Thinking-2507]; [Deepseek-ai/DeepSeek-R1]
Synthetic Science RL Data Blend from Qwen2.5-32B	Text	Undisclosed	[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 Tool Call Schema for RL	Text	469,983	[UltraTool]; [ToolEyes]; [AutoTools]; [API-Bank]; [Nemotron-Personas-USA]; [Salesforce xLAM function-calling]; [Glaive function-calling-v2]; [Agent-Ark/Toucan-1.5M]	[DeepSeek-V3.2]; [GLM-4.6]; [gpt-oss-120b]; [Kimi-K2-Instruct]
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	96,564	[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	Text	Undisclosed	[In-house indirect prompt injection data]	[nvidia/nemotron-3-super-v3, qwen/qwen3-next-80b-a3b-instruct]
Synthetic Malicious Code and Agentic Security	Text	Undisclosed	[In-house malicious-code / agentic-security data]	Undisclosed
Synthetic Single-Step SWE Patch Selection	Text	Undisclosed	[SWE-Gym Dataset]; [SWE Bench Verified Benchmark]	[ground truth and task checks]
Synthetic Natural Language Math Final Answers from Nemotron 5.5	Text	Undisclosed	[AMC8, AMC10, and AIME problem sets hosted on Art of Problem Solving]; [Pile-StackExchange]	[nemotron 5.5]
Synthetic Simple Math Prompts for Token Efficiency	Text	Undisclosed	[In-house simple math prompts]	Undisclosed
Synthetic Abstention Data from Nemotron Super v3 (CRAG)	Text	Undisclosed	[CRAG]	[nvidia/nvidia/nemotron-3-super-v3]
Synthetic Agentless SWE	Text	242,536	[SWE-Rebench-V2]; [SWEbench Training Set]; [R2E-Gym/R2E-Gym-Subset]; [SWE-Gym/SWE-Gym]; [SWE-Rebench]	[openai/gpt-oss-120b]
Synthetic Agentless SWE from DeepSeek-R1-0528	Text	209,976	[SWE-Bench-Train]; [SWE-Fixer-Train]; [SWE-reBench]; [SWE-Smith]	[deepseek-ai/DeepSeek-R1-0528]
Synthetic Agentic CUDA Traces from GLM-4.7	Text	2,276	[Internal CUDA task data]	[GLM-4.7]
Synthetic Math Proofs from DeepSeek-V3.2-Speciale	Text	820,772	[Nemotron-Math-Proofs-v1]	[SDG: DeepSeek-V3.2-Speciale]; [Filter: proof validation]
Synthetic Multilingual SFT from DeepSeek-V3	Text	1,245,284	[Nano v3 SFT data]	[DeepSeek-V3]
Synthetic Agentic Code from gpt-oss-120b	Text	109,086	[NVAgenticCLIPrompts-v1]; [NVAgenticSkills-v1]; [NVAgenticCLIMultiTurnPrompts-v1]	[openai/gpt-oss-120b]
Synthetic Agentic CLI and Web Skills from gpt-oss-120b	Text	27,418	[NVAgenticCLIPrompts-v1]; [NVAgenticSkills-v1]; [NVAgenticCLIMultiTurnPrompts-v1]; [NVAgenticCLIPrompts-Web-v1]	[openai/gpt-oss-120b]
Synthetic Agentic Coding from gpt-oss-120b	Text	160,531	[NVAgenticCLIPrompts-v1]; [NVAgenticSkills-v1]; [NVAgenticCLIMultiTurnPrompts-v1]; [NVAgenticCLIPrompts-Web-v1]	[openai/gpt-oss-120b]
Synthetic OpenCode Agentic Tasks from gpt-oss-120b	Text	614,773	[NVAgenticCLIPrompts-v1]; [NVAgenticSkills-v1]; [NVAgenticCLIMultiTurnPrompts-v1]; [NVAgenticCLIPrompts-Web-v1]	[openai/gpt-oss-120b]
Synthetic SWE Unverified	Text	Undisclosed	[NVAgenticCLIPrompts-v1]; [NVAgenticSkills-v1]; [NVAgenticCLIMultiTurnPrompts-v1]; [NVAgenticCLIPrompts-Web-v1]	[gpt-oss-120b]; [Qwen/Qwen3-Coder-480B-A35B-Instruct]; [GLM-4.7-Flash]
Synthetic ARC-AGI Ultra Data	Text	192,016	[ARC-AGI-2]; [arc dataset collection]	[ARC-AGI-2]
Synthetic LiveCodeBench TIR from DeepSeek-R1-0528	Text	1,283,398	[Nemotron-X training datasets]	[DeepSeek-R1-0528]
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	Text	Undisclosed	[OpenCodeReasoningv2]; [OpenMathReasoning]; [nemo-swe-bench-repos]; [SWE-Rebench]; [SWE-Fixer-110K]	[OpenReasoningv2]
Synthetic Tulu Instruction Following from DeepSeek-R1-0528	Text	105,361	[Nemotron-X training datasets]	[DeepSeek-R1-0528]
Synthetic Instruction Following from gpt-oss-120b	Text	151,988	[IFEval]; [IFEvalG]	[gpt-oss-120b]
Synthetic Instruction Following for RL	Text	Undisclosed	[WildChat-1M]; [LMSYS-340B-Eval Dataset]; [LMSYS-Chat-1M Prompts]; [IFEval]; [IFEvalG]	[Qwen/Qwen3-235B-A22B-Thinking-2507]; [gpt-oss-120b]; [Qwen3-235B-A22B-Instruct-2507]
Synthetic Identity Data from Qwen3-Next-80B-A3B-Instruct and Qwen3-235B-A22B-Instruct-2507	Text	25,992	[Hand-written prompts]	[Qwen3-Next-80B-A3B-Instruct]; [Qwen3-235B-A22B-Instruct-2507]
Synthetic Terminus Ultra Agentic Reasoning Blend	Text	96,881	[ARC-AGI-2]; [OpenCodeReasoningv2]; [OpenMathReasoning]; [SWE-Fixer-110K]; [SWE-Rebench]; [SWE-Smith]	[DeepSeek-V3.2]; [Qwen3-235B-A22B-Thinking-2507]; [Ring-1T]; [Kimi-K2.5]; [GLM-4.7-FP8]; [Qwen3-Next-80B-A3B-Thinking]; [gpt-oss-120b]; [Ministral-3-14B-Reasoning-2512]; [LM-4.5-Air-FP8]
Synthetic STEM from Qwen3-235B-A22B-Thinking-2507	Text	1,174,694	[IChO-IPhO-RL-v2]; [Physics-Big Dataset]; Scale HLE; [OpenMathReasoning]; [OpenCodeReasoning]	[Qwen3-235B-A22B-Thinking-2507]
Synthetic STEM from Qwen3-235B-A22B-Instruct-2507 and gpt-oss-120b	Text	Undisclosed	[arXiv]; [National Institutes of Health ExPorter]; [BioRxiv]; [PMC Article]; [USPTO Backgrounds]; [peS2o]; Global Regulation; [CORE]; [PG-19]; [DOAB CC BY & CC BY-SA subset]; [NDLTD]	[Qwen3-235B-A22B-Instruct-2507]; [gpt-oss-120b]
Translation Data from TAUS	Text	1,618,055	[TAUS proprietary dataset]	Undisclosed
Synthetic Art of Problem Solving and Stack Exchange from gpt-oss-120b, Qwen2.5-32B-Instruct, and Goedel-Prover-V2-32B	Text	860,469	[Nemotron-Math-Proofs-v1]	[Goedel-Prover-V2-32B]
Synthetic Art of Problem Solving and Stack Exchange from gpt-oss-120b	Text	1,201,815	[Upstream released math dataset]; [AoPS]; [StackOverflow / StackExchange]	[gpt-oss-120b]
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	Text	Undisclosed	[Nano-V3 SFT Data (without tool call)]	[Qwen/Qwen2.5-14B-Instruct]; [Qwen/Qwen3-4B-Thinking-2507]
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 (Stack Exchange lineage)	Text	Undisclosed	[Stack Exchange]; [SCP-116K]; [LIMO]; [TACO]; Code Contest; Codeforces	[DeepSeek-R1]; [DeepSeek-R1-0528]; [Qwen2.5-32B-Instruct]; [Qwen3-235B-A22B]
Synthetic Search Graph Walk	Text	6,977	[Wikidata / Wikipedia KnowledgeBase]	[MiniMaxAI/MiniMax-M2]
Synthetic Agentic Diverse Domains	Text	281,537	[Handwritten prompts (synthetic; no external seed data used)]	[SDG model: deepseek-ai/DeepSeek-V3.2, deepseek-ai/DeepSeek-R1-0528, Qwen/Qwen3-235B-A22B-Thinking-2507, Qwen/Qwen3-32B]; [Filtering model: openai/gpt-oss-120b, Qwen/Qwen3-32B, Qwen/Qwen3-235B-A22B-Instruct-2507]
Synthetic Long Context from Qwen3-235B-A22B-Instruct-2507	Text	Undisclosed	[Long-context SFT seed blend (pre-training blend + nano-v1 post-training data)]; [Long-context SFT data: lc_nothink 256k, MRCR 200k, RULER 256k]; [AALCR seed blend: SEC Filings, CC, Wikipedia, FinePDFs, ArXiv, Pile-NIH ExPorter, BioRxiv, PMC Article, USPTO Backgrounds, peS2o, Global Regulations, CORE, Gutenberg (PG-19), DOAB CC-BY, NDLTD, Amps, StackExchange, MathPile, Numinas]	[Qwen/Qwen3-235B-A22B-Thinking-2507]; [deepseek-ai/DeepSeek-R1]; [Qwen3-30B-A3B]
Synthetic Nemotron Math SFT from DeepSeek-V3.2-Speciale	Text	1,900,553	[Nemotron-Math-v2 (AOPS and StackExchange-math problems)]	[DeepSeek-V3.2-Speciale]
Synthetic Nemotron Math TIR from DeepSeek-V3.2	Text	1,789,258	[Nemotron-Math-v2 (AOPS and StackExchange-math problems)]	[DeepSeek-V3.2]
Synthetic NemoCascade OCR Distillation from gpt-oss-120b	Text	682,864	[Nemotron-X training datasets]	[gpt-oss-120b]
Synthetic CUDA 100k	Text	93,086	[KernelBook]; [HuggingFace Transformers]; [FlashInfer]	[gpt-oss-120b]; [DeepSeek-R1-0528]
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 Text-To-SQL from gpt-oss-120b	Text	Undisclosed	[In-house Text-to-SQL data]	[gpt-oss-120b]
Synthetic Tool Call Schema for RL (extended)	Text	707,967	[UltraTool]; [ToolEyes]; [AutoTools]; [API-Bank]; [Nemotron-Personas-USA]; [Salesforce xLAM function-calling]; [Glaive function-calling-v2]; [Agent-Ark/Toucan-1.5M]	[DeepSeek-V3.2]; [GLM-4.6]; [gpt-oss-120b]; [Kimi-K2-Instruct]
Synthetic Safety from gemma-3-4b-it, Nemotron-Nano-9B-v2, and gpt-oss-120b	Text	44,091	[Safety SFT Data]	[google/gemma-3-4b-it]; [Nemotron-Nano-9B-v2]; [gpt-oss-120b]
Synthetic Safety from DeepSeek-R1-0528, gpt-oss-120b, DeepSeek-R1-Distill-Qwen-7B, and Mixtral-8x7B-v0.1	Text	Undisclosed	[Nemotron Content Safety Dataset V2]; [Gretel Synthetic Safety Alignment Dataset]; [RedTeam-2K]; [Malicious Tasks]; [Nemotron-Personas-USA]	[DeepSeek-R1-0528]; [gpt-oss-120b]; [DeepSeek-R1-Distill-Qwen-7B]; [Qwen3-30B-A3B-Thinking-2507]; [Qwen3-235B-A22B-Instruct-2507]; [Mixtral-8x7B-v0.1]
Synthetic Tool Calling from Qwen3-235B-A22B-Thinking-2507 and Qwen3-Next-80B-A3B-Thinking	Text	Undisclosed	[ToolBench]; [glaive-function-calling-v2]; [APIGen Function-Calling]; [Nemotron-Personas-USA]	[Qwen3-235B-A22B-Thinking-2507]; [Qwen3-Next-80B-A3B-Thinking]
Synthetic Chat from gpt-oss-120b, Mixtral-8x22B-Instruct-v0.1, Qwen3-235B-A22B-Instruct-2507, and Qwen3-235B-A22B-Thinking-2507	Text	Undisclosed	[C4]; [LMSYS-Chat-1M]; [ShareGPT]; [GSM8K]; [PRM800K]; [FinQA]; [WikiTableQuestions]; [Riddles]; [glaive-function-calling-v2]; [SciBench]; [tigerbot-kaggle-leetcodesolutions-en-2k]; [OpenBookQA]; [Advanced Reasoning Benchmark]; Software Heritage; [Khan Academy Math Keywords]; [WildChat-1M]; [Nemotron-Personas-USA]	[gpt-oss-120b]; [Mixtral-8x22B-Instruct-v0.1]; [Qwen3-235B-A22B-Instruct-2507]; [Qwen3-235B-A22B-Thinking-2507]
Synthetic Tool Use Interactive Agent from gpt-oss-120b, DeepSeek-R1-0528, Qwen3-32B, and Qwen3-235B-A22B-Thinking-2507	Text	Undisclosed	NVIDIA Internal	[gpt-oss-120b]; [DeepSeek-R1-0528]; [Qwen3-32B]; [Qwen3-235B-A22B-Thinking-2507]
Synthetic DocFinQA and SWE-smith from Qwen3-Coder-480B-A35B-Instruct and Kimi-K2-Thinking	Text	Undisclosed	[DocFinQA]; [SWE-smith]	[Qwen3-Coder-480B-A35B-Instruct]; [Kimi-K2-Thinking]
Synthetic SWE-Gym from Qwen3-Coder-480B-A35B-Instruct	Text	Undisclosed	[SWE-Gym]	[Qwen3-Coder-480B-A35B-Instruct]
Synthetic SWE-Gym and R2E-Gym-Subset from Qwen3-Coder-480B-A35B-Instruct	Text	Undisclosed	[SWE-Gym]; [R2E-Gym-Subset]	[Qwen3-Coder-480B-A35B-Instruct]
Synthetic SWE-Gym and R2E-Gym-Subset from DeepSeek-R1-0528	Text	Undisclosed	[SWE-Gym]; [R2E-Gym-Subset]	[DeepSeek-R1-0528]
Synthetic HelpSteer, LMSYS-Chat-1M, and Nemotron-Personas-USA from gpt-oss-120b, Qwen3-235B-A22B-Instruct-2507, and Qwen3-235B-A22B-Thinking-2507	Text	Undisclosed	[HelpSteer2]; [HelpSteer3]; [LMSYS-Chat-1M]; [Nemotron-Personas-USA]	[gpt-oss-120b]; [Qwen3-235B-A22B-Instruct-2507]; [Qwen3-235B-A22B-Thinking-2507]
Synthetic Nemotron-Personas-USA from gpt-oss-120b and Qwen3-8B	Text	Undisclosed	[Nemotron-Personas-USA]	[gpt-oss-120b]; [Qwen3-8B]
Vendor Terminal Bench-like Tasks (Droid)	Text	Undisclosed	[Droid Harness Pivot vendor data]	Undisclosed
Synthetic BabyVisionQA Reasoning — Combined SFT	Text and images	135,207 records	Procedurally generated BabyVision-style visual-reasoning images	Qwen3.5-397B-A17B-FP8
Synthetic BabyVision Tool-Calling SFT, Relative-Path-Corrected Variant	Text and images	81,494 records	NVIDIA internal BabyVision tool-calling prompts and media	Qwen3.5-397B-A17B-FP8; Kimi-K2.6
Synthetic Chinese OCR v5, NMH5R	Text and images	29,108 records	Webster's Unabridged Dictionary; rendered Chinese text mixed with English words	Undisclosed; synthetic text rendering is described
Synthetic Chinese OCR v5, Augmented and Cleaned	Text and images	28,576 records	Webster's Unabridged Dictionary; augmented synthetic Chinese OCR images	Undisclosed; synthetic text rendering is described
Synthetic Multimodal Safety, Energon Format	Text and images	22,560 records	NVIDIA Safety-team text prompts and safe responses	Edify (image generation)
Synthetic Multimodal Safety, NMH5R and Safety-Filtered	Text and images	22,541 records	Multimodal safety records with Edify-generated images	Edify (image generation)
Synthetic ASCII OCR v5, NMH5R	Text and images	14,533 records	Webster's Unabridged Dictionary; synthetic ASCII renderings	Undisclosed; synthetic text rendering is described
Synthetic English OCR v5, NMH5R	Text and images	14,525 records	Webster's Unabridged Dictionary; synthetic English renderings	Undisclosed; synthetic text rendering is described

Language Distribution in Post-Training

For our post-training recipe, we focused on the following 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.

Additional Training Data

Data	Composition and Purpose
Indic continued pre-training	100B tokens: 10B selected Indic tokens and 90B tokens from the preceding blend.
Software-engineering mid-training	67B tokens: whole repositories (20%), issue–pull-request pairs (41%), commit diffs (3%), build and CI logs (1%), agent traces (10%), pre-training replay (20%), and long-context data (5%). Sources are decontaminated against SWE-bench and Terminal-Bench repositories.
Long-context pre-training	A 35B-token extension to 1M context, followed by a 9B-token long-context recovery stage.
Vision-language data	Image and video data from the Nemotron 3 Nano Omni training corpus, including visual question answering, charts, tables, OCR, multi-page documents, visual grounding, and video understanding.

Post-Training Data	Composition and Purpose
Additional synthetic vision data	Chart understanding, grounded perception, multimodal reasoning, and visual tool-use trajectories, including data generated with Qwen3.5 and Kimi models.
Reasoning trajectories	Mathematics and STEM responses generated with Kimi K3; mathematical proofs generated with DeepSeek V4 Pro; competitive-programming trajectories generated with GLM-5.2.
Chat and safety	Refreshed chat responses from Qwen, Kimi, and Gemma models, with additional safety data.
Long-context supervised data	512K-context data, synthetic graph reasoning and transduction tasks, long-horizon search and coding trajectories, and documents containing up to hundreds of pages.
Long-document translation	Chunked translations of PG19 and FinePDFs-Edu documents generated with DeepSeek V4 Flash.

Testing Datasets

Data Collection Method: Hybrid: automated, manually-collected, and synthetic
Labeling Method: Hybrid: automated, manually-labeled, and synthetic
Properties: Standard benchmarks and task-specific test suites covering reasoning, coding, tool use, instruction following, long context, and visual understanding.

Evaluation Datasets

Data Collection Method: Hybrid: automated, manually-collected, and synthetic
Labeling Method: Hybrid: automated, manually-labeled, and synthetic
Properties: Standard benchmarks and task-specific test suites covering reasoning, coding, tool use, instruction following, long context, and visual understanding.

Inference

• Acceleration Engine: PyTorch

• Test Hardware:

  • NVIDIA Hopper
    • H100
    • H200
  • NVIDIA Grace Blackwell
    • GB200
    • GB300
  • NVIDIA Blackwell
    • B200
    • B300

The deployment examples above provide configurations for NVIDIA Hopper and Blackwell systems. Choose the precision, context length, and concurrency appropriate for the application and available memory.

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.