Jua

What It Does

Jua is building a foundation model for physical reality, starting with the atmosphere. Unlike language models that learned the digital world from text, Jua's EPT-2 model learns physics directly from data, achieving state-of-the-art performance on atmospheric prediction against every incumbent, including ECMWF (the European Centre for Medium-Range Weather Forecasts), the institution that has defined weather forecasting for forty years.

The Platform: World Model + Agent

Jua combines two core components:

• EPT-2 World Model: A foundation model that learns physics from data, trained on the atmosphere but capable of transferring to other physical domains like airfoils and shock waves
• Athena Agent: An autonomous agent that resolves objectives inside physical reality by simulating consequences, calling tools, and optimizing outcomes

The agent and world model form a compounding loop where each improves the other over time.

Who It's For

Jua is currently in production with Fortune 500 companies, energy traders, utilities, and leading hedge funds. The platform powers over 100 GW of energy capacity worldwide, serving clients including Axpo, TotalEnergies, Shell, Enel, Statkraft, RWE, EDF, and others.

Key Capabilities

• State-of-the-art atmospheric prediction for energy grids, shipping lanes, and derivative markets
• Universal agent architecture that generalizes across physical objectives (prediction, trading, research acceleration)
• Physics transfer learning: the same base model, lightly fine-tuned, handles different physical domains
• Production-grade performance validated in real-world trading and operational environments

The Vision

Jua aims to become the foundation layer for physical AI by expanding beyond atmospheric modeling into turbomachinery, thermal systems, materials science, and any domain governed by physical equations. Their research is peer-reviewed at ICLR and NeurIPS, demonstrating rigorous scientific validation of their approach to learning continuous physics from data.