Jua EPT-2 AI Weather Model and Athena

Jua.ai AG is a Zurich-based research company that builds AI weather models for markets and operations that depend on the weather. Its team works with ETH Zurich researchers and is backed by the founder of The Climate Corporation and executives from DeepMind and Meta.

EPT-2 (Earth Physics Transformer) is a family of global, hourly AI weather models: deterministic, ensemble, rapid-refresh and a 5 km high-resolution model over Europe, refreshed up to 24 times a day with ensembles out to 60 days and wind and solar power forecasts in MW. Verified against more than 10,000 ground stations, it is reported to beat ECMWF HRES on wind, temperature and solar radiation. Athena is an AI agent that combines EPT-2 with market models, prices and positions to issue briefings and alerts. Customers include utilities, energy traders and hedge funds.

• EPT-2 global model: Hourly AI weather forecasts for wind, temperature and solar radiation worldwide.
• EPT-2e ensemble: Probabilistic forecasts out to 60 days that beat the 50-member ECMWF ENS mean.
• EPT-2 RR and HRRR: Rapid-refresh runs up to 24 times a day and a 5 km model over Europe.
• Power forecasts: Wind and solar output in MW for utility and renewable fleets, with actuals nowcasts.
• Athena agent: Reads forecasts, market models, prices and positions to issue briefings and alerts.
• API and Python SDK: Deterministic and ensemble data with decades of history for backtesting.

• A Vietnamese wind or solar farm operator forecasts MW output day-ahead and intraday to meet dispatch schedules.
• A power trading or retail unit uses temperature ensembles to anticipate demand peaks during heatwaves.
• A utility uses rapid-refresh forecasts to prepare grid operations before storms and strong wind events.

• More accurate wind, temperature and solar forecasts than ECMWF HRES across the 0 to 240 hour range, verified at over 10,000 stations.
• Up to 24 refreshes a day, so operators and traders react to changing weather faster than with 2 to 4 daily NWP runs.
• Converts weather forecasts directly into MW output for wind and solar fleets.
• Probabilistic ensembles out to 60 days for hedging, planning and backtesting.

AI foundation models are changing weather forecasting, delivering global forecasts faster and more often than traditional numerical models. As wind and solar take a larger share of power systems, including in Vietnam where renewable capacity has grown quickly and a competitive electricity market is developing, grid operators and generators need better short-term weather and output forecasts to balance supply and reduce curtailment.

Jua.ai AG is registered in Zurich, Switzerland (commercial register CHE-289.595.613), with its address at Lavaterstrasse 67. Founder and CEO Marvin Gabler. Research is published at NeurIPS and ICLR with ETH Zurich collaborators. The site lists energy companies such as TotalEnergies, Shell, Enel, Statkraft, RWE, EDF and Adani Energy among users of its models, powering over 100 GW.