- Cosmos 3 is the first open model to natively combine vision reasoning, world generation, and action prediction in a single mixture-of-transformers architecture.
- The model reduces physical AI training cycles from months to days by generating high-fidelity synthetic data for robotics and autonomous vehicles.
- Nvidia launched the Cosmos Coalition with Agile Robots, Runway, Skild AI, and others to advance open world models built on the Cosmos 3 foundation.
Cosmos 3 unifies reasoning and generation in a single physical AI architecture
At GTC Taipei on June 1, Nvidia announced Cosmos 3, what the company calls the world's first fully open omnimodel for physical AI. Unlike prior world foundation models that handled either vision understanding or video generation, Cosmos 3 does both in one system. Its mixture-of-transformers architecture pairs a reasoning transformer with an expert generation transformer, enabling the model to understand object interactions, motion, and spatial-temporal relationships before generating corresponding video and action trajectories.
The model can natively understand and generate text, images, video, ambient sound, and actions. According to Nvidia's official announcement, Cosmos 3 compresses physical AI training and evaluation cycles from months to days by generating synthetic training data with leading physics accuracy, meaning developers building autonomous vehicles or robotic systems can evaluate models against realistic simulated environments without requiring physical hardware.
Cosmos 3 is fully open, released with weights and documentation. The Cosmos 3 lineup includes Cosmos 3 Super for post-training robotics and AV models, Cosmos 3 Nano for high-speed action reasoning, and Cosmos 3 Edge (coming soon) for real-time inference at the edge.
"The big bang of physical AI is just around the corner thanks to breakthroughs in multimodal reasoning language, vision and world models. The Cosmos 3 family of open, frontier omnimodels gives developers a generational leap in ability to build robots, autonomous vehicles and vision AI that perceive, reason, plan and act in the physical world."
Jensen Huang, founder and CEO, NVIDIA — GTC Taipei keynote, June 1, 2026Why open physical AI lowers the cost of entry for robotics builders
Physical AI, the category covering systems that must perceive and act in the real world including robots, vehicles, and drones, has historically been bottlenecked not by algorithmic progress but by data. Collecting real-world training data for a robot is slow, expensive, and operationally difficult. The core promise of Cosmos 3 is that a robotics company no longer needs to own the full simulation stack. Instead, it can use a shared open foundation model to generate high-quality synthetic data, fine-tune on top of it, and focus engineering effort on its specific task domain.
Nvidia also launched the Cosmos Coalition, a collaborative initiative with AI labs and robotics leaders including Agile Robots, Black Forest Labs, Runway, and Skild AI. The structure mirrors how Meta's Llama releases created an ecosystem of fine-tuned variants across enterprise and research use cases, except here the downstream applications are physical systems rather than language applications. Physical AI developers already building on Cosmos 3 include Doosan Robotics, LG Electronics, Samsung, and Li Auto for autonomous vehicles.
The open release strategy positions Nvidia as the infrastructure layer beneath multiple competing robotics platforms, reinforcing its role as a picks-and-shovels player in the physical AI buildout regardless of which robotic applications ultimately win.
Cosmos 3 does not eliminate the data problem in physical AI. But it shifts who bears the cost of solving it, and that shift changes the competitive structure of every company building robots, autonomous vehicles, and industrial AI systems.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.