NVIDIA · 2026-07-15 · major
NVIDIA Cosmos 3 Edge — 4B world model that runs physical AI on Jetson
NVIDIA's smaller 4B-parameter Cosmos 3 world model runs on-device on Jetson, RTX GPUs, and DGX. It reasons about video, generates robot policies, and adapts to new robots in about a day.
NVIDIA's Cosmos family gets a small, on-device sibling built for real-time robotics.
Key specs
| Parameters | 4B |
|---|---|
| Adaptation time | ~1 day |
Quick facts
| Maker | NVIDIA |
|---|---|
| Parameters | 4B |
| Model family | Cosmos 3 (Nemotron backbone) |
| Hardware | Jetson T2000/T3000, RTX, DGX |
| Use cases | Robot policies, vision reasoning |
| Adaptation | ~1 day to new robot/sensor |
| Announced | 2026-07-15 (Tokyo) |
What is it?
Cosmos 3 Edge is a 4-billion-parameter world model that runs directly on robots, cameras, and vehicles instead of a data-center GPU. NVIDIA built it on the Nemotron backbone so it can perceive an environment, reason about it, and output the next policy step without a network round trip.
How does it work?
The model takes video and sensor input, updates its internal world model on-device, and produces action policies for the robot. NVIDIA says teams can fine-tune Cosmos 3 Edge to a specific robot, vehicle, or sensor setup in about a day, and target the new Jetson T2000 and T3000 modules alongside RTX GPUs and DGX systems.
Why does it matter?
Cosmos 3 Edge is aimed at the physical-AI stack that a factory arm or a delivery robot actually runs. Cutting the world model down to 4B parameters lets it live on-device with real-time latency, so robotics teams no longer need a cloud call to get the next action.
Who is it for?
robotics and physical-AI teams
Frequently asked questions
- How is Cosmos 3 Edge different from the original Cosmos 3?
- Cosmos 3 Edge is the on-device sibling of Cosmos 3, at 4 billion parameters instead of the Nano (16B) and Super (65B) variants NVIDIA showed at Computex. It targets Jetson, RTX, and DGX systems so robots can run world-model reasoning locally in real time rather than calling a data-center GPU.
- What hardware does Cosmos 3 Edge run on?
- Cosmos 3 Edge is designed for NVIDIA Jetson, including the newly announced T2000 and T3000 modules, as well as RTX GPUs and DGX systems. That range lets a factory robot, a delivery vehicle, or a workstation-based simulation share the same model weights.
- How long does it take to adapt Cosmos 3 Edge to a specific robot?
- NVIDIA says developers can adapt Cosmos 3 Edge to a specific robot, vehicle, sensor set, or environment in about a day. The open framework around Cosmos is designed for this kind of per-deployment fine-tuning without retraining the base world model.
- Which Japanese companies are joining the Cosmos Coalition?
- The July 16 announcement in Tokyo added AIRoA, FANUC, Fujitsu, Hitachi, Honda R&D, Kawasaki Heavy Industries, Kubota, NEC, SoftBank, Sony, Yaskawa Electric, and others to the NVIDIA Cosmos Coalition. The group commits to building on top of the open Cosmos frontier for physical AI.
Try it
Available through NVIDIA developer channels; new Jetson T2000/T3000 modules ship alongside.