Uploaded June 2026 | Updated September 2026, 2 weeks ago
In this tutorial, you will learn how to pull and run the NVIDIA Cosmos 3 Reasoner NIM from NGC and query it with your own videos and prompts. The Cosmos 3 Reasoner NIM is a pre-packaged container for running Cosmos 3 vision language model (VLM) inference — you give it a text prompt and a video or image, and it returns detailed reasoning responses. By the end, you will have the Cosmos 3 Reasoner NIM running locally and be able to send inference requests across robotics, autonomous driving, and data curation use cases.
Get started with Cosmos 3:
📘 Read the technical blog → nvda.ws/4u5G0x8
🤗 Download models → huggingface.co/collections/nvidia/cosmos3
⚙️ Customize + post-train → github.com/nvidia/Cosmos
🚀 Deploy Cosmos 3 models → build.nvidia.com/models?q=cosmos
Chapters:
00:00 Intro — What is the Cosmos 3 Reasoner NIM?
00:30 Find the Cosmos 3 NIM container on NGC Catalog
01:00 Set your NGC API key and log in to Docker
01:30 Pull and launch the NIM container
02:00 Check health and readiness endpoints
02:30 Pass images to the API
03:00 Run your first inference request
03:30 Pass videos to the API
04:30 Example: Synthetic data generation critic (Python SDK)
05:30 Run inference and review model reasoning output
06:15 Wrap-up and next steps
#NVIDIACosmos #WorldFoundationModels #PhysicalAI
NVIDIA Cosmos, World Foundation Models, Physical AI, Cosmos 3 Reasoner NIM
Q: What is the NVIDIA Cosmos 3 Reasoner NIM?
A: It is a pre-packaged container that lets you run Cosmos 3 vision language model inference locally via API — you send a text prompt and a video or image, and it returns a detailed reasoning response.
Q: How long does it take to get the Cosmos 3 Reasoner NIM running?
A: After setting your NGC API key and running the docker pull command, the container takes roughly 2 to 5 minutes to launch on first run. Subsequent launches are faster.
Q: What GPU do I need to run the Cosmos 3 Reasoner NIM?
A: You can run the Cosmos 3 Reasoner NIM on a DGX Spark, RTX PRO 6000 GPU, Hopper (H100, H200), or Blackwell GPUs (B200, B300). At least 30GB free disk space is recommended.
Q: Can I use my own videos with the Cosmos 3 Reasoner NIM?
A: Yes. You can replace the video URL in the API request with any publicly accessible video link and modify the text prompt to ask whatever you need.
Q: Where can I find more API examples?
A: Full documentation and additional API usage examples are in the NVIDIA Cosmos docs. The GitHub repo at github.com/nvidia/Cosmos also has recipes and post-training guides.
In this tutorial, you will learn how to pull and run the NVIDIA Cosmos 3 Reasoner NIM from NGC and query it with your own videos and prompts. The Cosmos 3 Reasoner NIM is a pre-packaged container for running Cosmos 3 vision language model (VLM) inference — you give it a text prompt and a video or image, and it returns detailed reasoning responses. By the end, you will have the Cosmos 3 Reasoner NIM running locally and be able to send inference requests across robotics, autonomous driving, and data curation use cases.
Get started with Cosmos 3:
📘 Read the technical blog → nvda.ws/4u5G0x8
🤗 Download models → huggingface.co/collections/nvidia/cosmos3
⚙️ Customize + post-train → github.com/nvidia/Cosmos
🚀 Deploy Cosmos 3 models → build.nvidia.com/models?q=cosmos
Chapters:
00:00 Intro — What is the Cosmos 3 Reasoner NIM?
00:30 Find the Cosmos 3 NIM container on NGC Catalog
01:00 Set your NGC API key and log in to Docker
01:30 Pull and launch the NIM container
02:00 Check health and readiness endpoints
02:30 Pass images to the API
03:00 Run your first inference request
03:30 Pass videos to the API
04:30 Example: Synthetic data generation critic (Python SDK)
05:30 Run inference and review model reasoning output
06:15 Wrap-up and next steps
#NVIDIACosmos #WorldFoundationModels #PhysicalAI
NVIDIA Cosmos, World Foundation Models, Physical AI, Cosmos 3 Reasoner NIM
Q: What is the NVIDIA Cosmos 3 Reasoner NIM?
A: It is a pre-packaged container that lets you run Cosmos 3 vision language model inference locally via API — you send a text prompt and a video or image, and it returns a detailed reasoning response.
Q: How long does it take to get the Cosmos 3 Reasoner NIM running?
A: After setting your NGC API key and running the docker pull command, the container takes roughly 2 to 5 minutes to launch on first run. Subsequent launches are faster.
Q: What GPU do I need to run the Cosmos 3 Reasoner NIM?
A: You can run the Cosmos 3 Reasoner NIM on a DGX Spark, RTX PRO 6000 GPU, Hopper (H100, H200), or Blackwell GPUs (B200, B300). At least 30GB free disk space is recommended.
Q: Can I use my own videos with the Cosmos 3 Reasoner NIM?
A: Yes. You can replace the video URL in the API request with any publicly accessible video link and modify the text prompt to ask whatever you need.
Q: Where can I find more API examples?
A: Full documentation and additional API usage examples are in the NVIDIA Cosmos docs. The GitHub repo at github.com/nvidia/Cosmos also has recipes and post-training guides.
