At Jua, our vision is to build a single model of our physical world, a foundation world model. You can think of it like sparse video prediction: we take in a stream of observations from physical sensors across the planet and learn to predict how the entire system evolves. Assimilation, bias correction, time integration, and cross-modality generalization emerge naturally within a unified architecture.
This approach is still early, but every iteration has already surpassed existing systems. Starting with weather, our models consistently outperform the most advanced forecasts in accuracy, speed, and adaptability. The foundation extends far beyond weather: we're beginning to see generalization to climate scales, wildfire dynamics, energy markets, and even fundamental fluid and thermodynamic problems. The model is learning patterns and physical structures that were once thought too complex to capture directly from data.
We believe this is how humanity will eventually master Earth prediction, not with more fragmented tools, but with a unified, ever-improving model of the planet itself.
While the majority of our code is closed source yet, we open source various modules & helper tools. It is planned to open source past model versions and eventually provide access to a non-commercial community version of the current model.
There are various ways to get involved:
- Submit pull requests
- Drop us a message @ [email protected]
- Sign up to our newsletter and follow us on our Social Media channels to stay updated
- Spread the word