Deployment Geography:

Global

Use Case:

Robotics engineers and AI researchers developing embodied agents—such as autonomous vehicles and robotic systems—would use Physical AI to equip their machines with spatiotemporal reasoning and a fundamental physics understanding for navigation, manipulation, and decision-making tasks Reference: Technical Paper

Release Date:

Build.NVIDIA.com 08/11/2025 via link
Huggingface 08/01/2025 via link

Model Architecture:

Architecture Type: A Multi-modal LLM consists of a Vision Transformer (ViT) for vision encoder and a Dense Transformer model for LLM.\ Network Architecture: Qwen2.5-VL-7B-Instruct.\ Cosmos-Reason-7B is post-trained based on Qwen2.5-VL-7B-Instruct and follows the same model architecture. Number of model parameters: Cosmos-Reason1-7B:

• Vision Transformer (ViT): 675.76M (675,759,104)
• Language Model (LLM): 7.07B (7,070,619,136)
• Other components (output projection layer): 545.00M (544,997,376)

Computational Load:

• Cumulative Compute: 3.2603016e+21 FLOPS
• Estimated Energy and Emissions for Model Training:
• Total kWh = 16658432
• Total Emissions (tCO2e) = 5380.674

Input:

Input Type: Text, Image
Input Parameters: Text: One Dimensional (1D), Video (3D)
Input Format: Text, Video
Other Properties Related to Input:

• Use FPS=4 for input video to match the training setup.
• Append
  Answer the question in the following format: \nyour reasoning\n\n\n\nyour answer\n. in the system prompt to encourage long chain-of-thought reasoning response.

Input Context Length (ISL): 128K

Output:

Output Type: Text
Output Parameters: One Dimensional (1D)
Output Format: String
Other Output Properties: Recommend using 4096 or more output max tokens to avoid truncation of long chain-of-thought response. Our AI model recognizes timestamps added at the bottom of each frame for accurate temporal localization. Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.__

Software Integration:

Runtime Engines:

• vLLM

Supported Hardware Platforms:

• NVIDIA Ampere
• NVIDIA Hopper

Training, Testing, and Evaluation Datasets:

05/17/2025 Please see our technical paper for detailed evaluations on physical common sense and embodied reasoning. Part of the evaluation datasets are released under Cosmos-Reason1-Benchmark . The embodied reasoning datasets and benchmarks focus on the following areas: robotics (RoboVQA, BridgeDataV2, Agibot, RobFail), ego-centric human demonstration (HoloAssist), and Autonomous Vehicle (AV) driving video data. The AV dataset is collected and annotated by NVIDIA. All datasets go through the data annotation process described in the technical paper to prepare training and evaluation data and annotations. 08/01/2025 We enhance the model capability with the augmented training data. PLM-Video-Human and Nexar are used to enable dense temporal captioning. Describe Anything is added to enhance a set of mark (SoM) prompting. We enrich data in intelligent transportation systems (ITS) and warehouse applications. Lastly, Visual Critics dataset contains a collection of AI generated videos from Cosmos-Predict2 and Wan2.1 with human annotations to describe the physical correctness in AI videos.

Training Dataset:

• Data Modalities: Video, Text
• Text Training Data Size: Not specified (included in multimodal inputs)
• Video Training Data Size: Less than 10,000 Hours (~2M video-text pairs in SFT + RL datasets)

Data Collection Method by dataset:

• RoboVQA: Hybrid: Automatic/Sensors
• BridgeDataV2: Automatic/Sensors
• AgiBot: Automatic/Sensors
• RoboFail: Automatic/Sensors
• HoloAssist: Human
• AV: Automatic/Sensors
• PLM-Video-Human: Human
• Nexar: Automatic/Sensors
• Describe Anything: Human
• ITS / Warehouse: Human, Automatic
• Visual Critics: Automatic

Labeling Method by dataset:

• RoboVQA: Hybrid: Human,Automated
• BridgeDataV2: Hybrid: Human,Automated
• AgiBot: Hybrid: Human,Automated
• RoboFail: Hybrid: Human,Automated
• HoloAssist: Hybrid: Human,Automated
• AV: Hybrid: Human,Automated
• PLM-Video-Human: Human,Automated
• Nexar: Human
• Describe Anything: Human,Automated
• ITS / Warehouse: Human, Automated
• Visual Critics: Human,Automated

Evaluation Dataset:

• RoboVQA: Hybrid: Automatic/Sensors
• BridgeDataV2: Automatic/Sensors
• AgiBot: Automatic/Sensors
• RoboFail: Automatic/Sensors
• HoloAssist: Human
• AV: Automatic/Sensors

Labeling Method by dataset:

• RoboVQA: Hybrid: Human,Automated
• BridgeDataV2: Hybrid: Human,Automated
• AgiBot: Hybrid: Human,Automated
• RoboFail: Hybrid: Human,Automated
• HoloAssist: Hybrid: Human,Automated
• AV: Hybrid: Human,Automated

Evaluation Benchmark Results:

We report the model accuracy on the embodied reasoning benchmark introduced in Cosmos-Reason1. The results differ from those presented in Table 9 due to additional training aimed at supporting a broader range of Physical AI tasks beyond the benchmark.

Evaluation Results

Dataset	RoboVQA	AV	BridgeDataV2	Agibot	HoloAssist	RoboFail	Average
Accuracy	87.3	70.8	63.7	48.9	62.7	57.2	65.1

Dataset Format

Modality: Video (mp4) and Text

Dataset Quantification

05/17/2025 We release the embodied reasoning data and benchmarks. Each data sample is a pair of video and text. The text annotations include understanding and reasoning annotations described in the Cosmos-Reason1 paper. Each video may have multiple text annotations. The quantity of the video and text pairs is described in the table below. The AV data is currently unavailable and will be uploaded soon!

Dataset	RoboVQA	AV	BridgeDataV2	Agibot	HoloAssist	RoboFail	Storage Size
SFT Data	1.14m	24.7k	258k	38.9k	273k	N/A	300.6GB
RL Data	252	200	240	200	200	N/A	2.6GB
Benchmark Data	110	100	100	100	100	100	1.5GB

We release text annotations for all embodied reasoning datasets and videos for RoboVQA and AV datasets. For other datasets, users may download the source videos from the original data source and find corresponding video sources via the video names. The held-out RoboFail benchmark is released for measuring the generalization capability.

Inference:

Test Hardware: H100
Note: We suggest using fps=4 for the input video and max_tokens=4096 to avoid truncated response.

Python from transformers import AutoProcessor
from vllm import LLM, SamplingParams
from qwen_vl_utils import process_vision_info\ #You can also replace the MODEL_PATH by a safetensors folder path mentioned above
MODEL_PATH = "nvidia/Cosmos-Reason1-7B"\ llm = LLM(
>model=MODEL_PATH,
limit_mm_per_prompt={"image": 10, "video": 10},
)

Sampling Parameters Configuration

Python sampling_params = SamplingParams(\

temperature=0.6,
top_p=0.95,
repetition_penalty=1.05,
max_tokens=4096,
)

Video Message Setup

Python
video_messages = [\

{"role": "system", "content": "You are a helpful assistant. Answer the question in the following format: \nyour reasoning\n\n\n\nyour answer\n."},
{"role": "user", "content": [
{"type": "text", "text": (
"Is it safe to turn right?"
)
},
{
"type": "video_url",\
"video_url": {"url": BASE_64_ENCODED_VIDEO},
"fps": 4,
}
]
},
]

Key Inference Notes

• FPS Requirement: Use FPS=4 for input video to match the training setup
• System Prompt: Append "Answer the question in the following format: \nyour reasoning\n\n\n\nyour answer\n." in the system prompt to encourage long chain-of-thought reasoning response
• Output Tokens: Recommend using 4096 or more output max tokens to avoid truncation of long chain-of-thought response
• Hardware: Model is designed and optimized to run on NVIDIA GPU-accelerated systems
• Precision: Tested with BF16 precision for inference
• Operating System: Tested on Linux (other operating systems not tested)

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

NVIDIA Cosmos Reason NIM is optimized to run best on the following compute:

GPU	Total GPU memory	Azure VM compute	#GPUs on VM	Link
A100	160	Standard_NC48ads_A100_v4	2	link
H100	94	STANDARD_NC40ADS_H100_V5	1	link