A foundation
model for reality.

Language models learned the digital world from text. The physical world is larger, older, and governed by different laws. It needs a different model.

The thesis

The last three years changed everything you do on a screen. The next three will change everything else.

The medicines in your cabinet. The plane you flew on last month. The power that just turned on this light. The food in your fridge, the car in your driveway, the chip inside your phone. Every one of these comes from a world AI has barely touched. A world where design cycles are measured in years. Where a single experiment can cost a billion dollars. Where most ideas die because nobody could afford to try them.

LLMs can't fix this. What the physical world needs is what language needed: a foundation model of how reality actually behaves, and an agent that acts inside it. That is what Jua is building. We started with the atmosphere, because if a model can learn coupled fluid dynamics, thermodynamics, and radiative transfer at planetary scale, it can learn anything. It did. Now we go everywhere.

Powering 100+ GW worldwide
Axpo
TotalEnergies
Shell
Enel
Statkraft
EnBW
EDF
Hydro-Québec
Adani Energy
Vitol
Origin Energy
ESB
01 · World model
EPT-2

A foundation model that learns physics from data.

State of the art on atmospheric prediction against every incumbent, including ECMWF, the institution that has defined the field for forty years. Trained on weather. The same base, lightly finetuned, then handled airfoils and shock waves. The physics transferred. The domain is a variable.

02 · Agent
Athena

An agent that resolves objectives inside physical reality.

Give it an objective, a world model, and a set of tools. It simulates consequences, calls the tools, and resolves. In production in utilities, energy traders and hedge funds.

The compounding loop

The agent improves the world model. The world model makes the agent more capable.

World modelEPT-2AgentAthenaObjectivedata · toolssimulates consequencesresolvesimproves
Where we started

The atmosphere is the hardest continuous-physics dataset humanity has ever recorded.

A chaotic, high-dimensional, partially observed fluid, recorded at kilometer resolution for decades. If a model can learn the physics of the atmosphere from data, it can learn the physics of almost anything. Our research is peer-reviewed at ICLR and NeurIPS.

Atmospheric prediction · aggregate skill
EPT-2Jua
100
GenCastGoogle DeepMind
84
AuroraMicrosoft
79
FourCastNet 3NVIDIA
73
IFSECMWF
68
Aggregate skill across RMSE, ACC, and CRPS on a held-out 2024 test set, normalized to EPT-2 = 100. Illustrative. Full tables in the EPT-2 technical report.
Three objectives, one agent

The agent is universal. The objective is the variable.

Language models generalized across text tasks. An agent that resolves physical objectives should generalize across physical objectives. We wanted to test that.

I
Objective · atmospheric prediction

Minimize forecast error. Win against ECMWF.

Energy grids, shipping lanes, and derivative markets depend on knowing what the atmosphere will do. Athena, pointed at this objective, beats every incumbent forecast system on the metrics traders actually price.

II
Objective · prediction-market alpha

Price a contract better than the market prices it.

Athena runs the Jua employee quant fund. Same agent. Different tools. A different loss function. It made money from day one.

III
Objective · accelerate research

Help the team that builds the next model.

Pointed at our own evaluation metric, Athena helps the research team set up experiments, pull data, and inspect results on our GPU cluster. Humans still drive it; the loop is getting shorter.

The sequence

In this order.

Each step funds and hardens the next. None of them is optional.

01

Build the best atmospheric model in the world.

Prove the foundation model on the hardest continuous-physics dataset on Earth.
02

Point the agent at objectives that matter.

Inside the atmosphere first, because that is where we can show it works today.
03

Expand the world model beyond the atmosphere.

Turbomachinery. Thermal. Materials. Any domain that obeys a governing equation.
04

Become the layer every physical AI system runs on top of.

Foundation models did this for language. Jua does it for reality.

We are building the foundation layer for physical AI.

A foundation model of how reality behaves. An agent that acts inside it. In production globally in fortune 500 companies, energy traders and leading hedge funds.