Roboflow joins NVIDIA Cosmos Coalition, adds Cosmos 3 post-training

What the coalition is and why Roboflow joined

According to Roboflow, NVIDIA introduced the Cosmos Coalition in May 2026 to bring together model builders, developers and organizations working on physical AI around open world models. The coalition supports collaboration on models, research and evaluation methods, with the aim of advancing systems that can understand and simulate the physical world.

Roboflow says its membership expands its existing collaboration with NVIDIA and the work of Roboflow Labs, its research arm. The company describes the move as a way to place that work within a shared effort to advance real-world capabilities.

Roboflow argues that the next wave of the AI boom will be defined by what intelligence can do in the physical economy. It points to factories that make products, infrastructure that connects cities, and the machines people work alongside every day. In its view, bringing AI into those environments could make capabilities that once required custom engineering accessible to many more teams.

Research shaped by real developer tasks

Roboflow Labs draws on what the company calls the "2M+ developers and Fortune 500 enterprises building on Roboflow". According to Roboflow, that real-world task distribution shapes the research, the data, and benchmarks such as RF100-VL that it brings to model builders.

The company says it sees an opportunity to connect progress in frontier models with the breadth of tasks developers face outside the lab, and to help turn that progress into tools people can build with.

Cosmos 3 post-training on the Roboflow platform

The practical part of the announcement is that NVIDIA Cosmos 3 post-training is now available on Roboflow. It starts with video reasoning. The source does not give pricing, regional availability or technical specifications.

Post-training lets developers adapt a pretrained model to a specific task or environment. Roboflow gives the example of interpreting how work progresses through an assembly process. The company calls this one practical expression of the broader Labs mission: helping advanced models become more capable in the settings where people need them.

What it means for developers

For teams building computer vision and physical AI systems, the change is that a Cosmos 3 model can be adapted to their own task or environment through Roboflow, beginning with video reasoning use cases. Roboflow says its ambition is for visual intelligence to become foundational infrastructure for the physical economy, and that it is pursuing a future in which developers can bring intelligence to the world around them as readily as they build software today.

Roboflow invites anyone building models that need to understand more of the physical world to get in touch with Roboflow Labs.