NVIDIA introduces Cosmos 3 Edge for on-device robot and vision-AI reasoning
NVIDIA has introduced Cosmos 3 Edge, a version of its Cosmos world-model family built to run reasoning directly on edge hardware rather than in the cloud, alongside a wider push into Japan's physical-AI and robotics sector.
What's new
NVIDIA describes Cosmos 3 Edge as enabling "on-device vision reasoning and robot policy deployment on NVIDIA Jetson Thor platforms." Unlike cloud-hosted world models that reason about an environment remotely, Cosmos 3 Edge is designed to run its perception and decision-making locally on the robot or device itself — letting a machine interpret its surroundings and generate actions without a round trip to a data center. NVIDIA paired the release with new Metropolis libraries, also built on Cosmos, aimed at agentic vision-AI development.
The release came alongside a broader announcement that Japan's physical-AI and manufacturing leaders — including AIRoA, FANUC, Fujitsu, Hitachi, Kawasaki Heavy Industries, Kubota, NEC, SoftBank, Sony, and Yaskawa Electric — intend to join the NVIDIA Cosmos Coalition, an industry group working to build open frontier physical-AI models on the Cosmos platform.
Context
Cosmos 3 Edge fills in a piece NVIDIA had flagged as "coming soon" when it first launched the broader Cosmos 3 lineup earlier this year. Edge deployment has been the harder half of the physical-AI pitch: a world model that reasons well is only useful for a factory robot or autonomous system if it can run fast enough, and locally enough, to act in real time. Pairing the model release with the Cosmos Coalition membership push also reflects NVIDIA's platform strategy in physical AI — recruiting manufacturers and robotics integrators as ecosystem partners who build on Cosmos rather than competing model stacks.
Why it matters
Edge-deployable reasoning is the piece that turns a world model from a simulation and training tool into something that can actually run inside a factory robot, warehouse vehicle, or piece of construction equipment without depending on network connectivity or cloud latency. Coming during NVIDIA's push into Japan's robotics and manufacturing base — with a long list of established industrial names lined up to build on the platform — Cosmos 3 Edge is as much a signal of where NVIDIA expects the next phase of physical-AI competition to play out (on-device, in real industrial settings) as it is a standalone model release.
Corroborating sources
- Nvidianews.nvidia
https://nvidianews.nvidia.com/news/japans-robotics-and-manufacturing-leaders-build-on-nvidia-cosmos-to-advance-physical-ai-frontier
“NVIDIA introduces Cosmos 3 Edge for on-device vision reasoning and robot policy deployment on NVIDIA Jetson Thor platforms”
- Siliconangle
https://siliconangle.com/2026/07/16/nvidia-launches-cosmos-3-edge-model-expands-physical-ai-push-japan/