NVIDIA
Explore
Models
Blueprints
GPUs
Docs
⌘KCtrl+K
Terms of Use
Privacy Policy
Your Privacy Choices
Contact

Copyright © 2025 NVIDIA Corporation

nvidia

cosmos-reason1-7b

Run Anywhere

Reasoning vision language model (VLM) for physical AI and robotics.

Physical AIautonomous vehiclesindustrialreasoningroboticssmart citiesSynthetic Data Generationvideo understandingvision language model
Download Now

Input

Drop files here.mp4
.mp4
Your question or task. Aim 20-150 tokens (~15-115 words); max 250 tokens.
Defines AI role/rules for session. Aim 100-400 tokens (~75-300 words); max 1,000 tokens. Model can accommodate reasoning or non-reasoning answers. Enable reasoning by including this text string in the system prompt:<think>your reasoning</think><answer>your answer</answer>
Using free API for development

You may upload videos for the Cosmos Reason1-7B trial experience. NVIDIA will only use and store those videos to provide you with this trial experience. Please do not upload videos containing confidential information or personal data. Your use is logged for security, fraud or abuse monitoring. For more information about our data processing practices, see our Privacy Policy. By clicking “Run” you consent to our collection and use of uploaded videos and you agree to the NVIDIA API Trial Service Terms of Use. Your use of the model is governed by NVIDIA Open Models License.

Output

Reasoning Complete

Below is the entire thinking process the model went through to arrive at its response.

Response