NVIDIA Cosmos Cookoff
Host: NVIDIA
Sponsors: Nebius and Milestone Systems
Community: Discord
Prizes:
First Place: $3,000 and an NVIDIA DGX Spark™
Second Place: $2,000 and an NVIDIA GeForce RTX™ 5090 GPU
Third Place: $500 brev credits
Overview
Join the Cosmos™ Cookoff, a hands-on physical AI challenge for developers, researchers, and system builders in robotics, autonomous vehicles, and vision AI. Build innovative solutions with Cosmos Reason 2, a leaderboard-topping, open reasoning vision language model that helps AI agents and autonomous systems see, understand, and act in the physical world like humans.
Why Participate?
This is an opportunity to level up your skills with cutting-edge physical AI models, build a project to solve a real-world problem, connect with experts, and learn alongside a community of physical AI enthusiasts. It’s your chance to show the world what’s possible with Cosmos.
The Challenge
Build something great with Cosmos Reason 2—post-train for a specialized reasoning model, a data curation tool, a robot brain that understands a new domain, or a video analytics agent.
Get creative with Cosmos Reason. You can use it standalone or mix in other open models to bring your vision to life.
Access NVIDIA experts, who will provide guidance, feedback, and technical help through office hours and our Discord channel to ensure you make the most of Cosmos Reason.
Note: Cosmos Reason is purpose-built for physical AI. This challenge does not cover generative media, art, gaming, or content creation projects.
Who Should Participate
Developers and researchers building robotics, autonomous vehicles, and vision AI pipelines
Teams (1-4 members)
Those looking to experiment, evaluate, and benchmark physical AI workflows in a safe, supported environment
Additional Tools From the Cosmos Cookbook
Kick off with Cosmos Cookbook recipes, but you can also adapt, extend, or innovate across robotics, autonomy, simulation, transportation, or more. We recommend the following recipes:
Egocentric Social and Physical Reasoning—Test and evaluate robot-human social and spatial interactions with reasoning.
Post-Training Cosmos Reason (Intelligent Transportation)—Learn how to post-train Cosmos Reason using a supervised fine-tuning technique with an example transportation dataset to better understand traffic scenarios.
Physical Plausibility Prediction—Synthetic data generation is a game-changer, but not all of the outputs are usable. This recipe shows how to evaluate the physical plausibility of synthetically generated videos.
Build a Video Analytics AI Agent—Unlock the power of agentic AI to understand, reason, and act on your live video streams or stored files. Use the NVIDIA Blueprint for video search and summarization, and start getting insights.
Schedule
Competition Schedule
January 29, 9 a.m. PT—Registration opens
February 19, 9 a.m. PT—Registration closes
February 26, 5 p.m. PT—Submission closes
Week of March 16—Winners announced
Activity Schedule
February 5, 9 a.m. PT—Tutorial Livestream: Build Reasoning Agents For Physical AI
February 10, 9 a.m. PT—Livestream: Intro to NVIDIA Cosmos | With Ming-Yu ft. Superintelligence
February 10, 10 a.m. PT—Ask Me Anything
February 17, 9 a.m. PT—Ask Me Anything
February 19, 9 a.m. PT—Partner Livestream: Physical AI in Action With Cosmos Reason
February 24, 9 a.m. PT—Ask Me Anything
How to Participate
Registration opens: January 29
Have a valid submission idea ready to go.
Registration closes: February 19 to be eligible for prizes
Join the community: NVIDIA Developer Discord for announcements, discussions, and expert support
Submit your project here. Be sure to include:
Text description that explains the features and functionality of your project
Demo video (<3minutes)
URL to your public code repository
README file that contains deployment instructions
Submission deadline for prizes: February 26
Scoring and Judging Criteria
Quality of the Ideas*
Does this solution demonstrate a compelling application of Cosmos Reason for robotics, autonomous systems, and video analytics AI agents?
Technical Implementation
Is the software development high quality and easy to follow, reproduce, and evaluate?
Is there clear documentation of inputs, outputs, and evaluation steps?
Design
Is the design of the project and user experience well thought out and intuitive?
Impact
What problem does this solve?
How does it move the physical AI field forward and/or impact the world?
Judges
Your projects will be reviewed by experts from Datature, Hugging Face, Nebius, Nexar, and NVIDIA, bringing deep experience in open models, cloud/compute, and real-world edge and vision AI deployments.
*Fairness and Reviewer Consistency
We ensure fairness through a diverse judging committee composed of both internal and external experts from different backgrounds. All judges will receive standardized training and calibration guidelines to promote consistent scoring and reduce individual bias.
Resources and Legal
Learning: NVIDIA Cosmos Cookoff—Participation Guide
By filling out the form, you agree to share your data with our partners, Milestone and Nebius. Your information will be handled in accordance with Milestone’s Privacy Policy and Nebius’s Privacy Policy.