Jua’s New AI Weather Model

Swiss startup Jua announced the release of its AI‑based weather model, EPT‑2, asserting that it delivers the most accurate forecasts across a range of metrics. The company’s claim rests on a newly published report that pits EPT‑2 against several top‑tier forecasting systems, including Microsoft’s Aurora, the European Centre for Medium‑Range Weather Forecasts (ECMWF) ENS model, and the IFS HRES model.

Performance Compared to Competitors

According to the report, EPT‑2 outperformed Aurora on key variables such as 10‑metre wind speed and 2‑metre air temperature over a ten‑day forecast horizon. The model also ran forecasts 25% faster than Aurora and recorded the lowest error scores among all models tested. In addition, Jua states that EPT‑2 achieved these results while consuming 75% less computing power than Aurora, making it the most efficient system in the study.

Study Scope and Limitations

The comparative analysis included two of ECMWF’s best‑known models—ENS and IFS HRES—but did not incorporate Google DeepMind’s Graphcast model. Despite this omission, Jua’s CEO and co‑founder Marvin Gabler expressed confidence that EPT‑2 could also surpass Graphcast, noting respect for existing players while highlighting perceived drawbacks such as speed, scope, and reliance on legacy infrastructure.

AI in Weather Forecasting

AI‑based forecasting has gained traction as a means to deliver more accurate and cost‑effective predictions. Traditional models rely on complex physics equations run on expensive supercomputers, whereas AI models learn patterns from massive datasets, potentially delivering forecasts thousands of times faster on far cheaper hardware. Jua claims to go a step further by constructing a native physics simulation that directly captures atmospheric behavior, rather than merely attaching AI layers to existing legacy systems.

Company Background and Funding

Jua released its first global AI weather model three years ago and has since secured $27 million in funding from investors including 468 Capital, Future Energy Ventures, and Promus Ventures. The company positions itself as a challenger to established players, aiming to combine speed, accuracy, and computational efficiency in a single forecasting solution.

Future Publication

The research underpinning Jua’s claims is slated for publication on the open‑access archive arXiv in the coming week, providing the broader scientific community an opportunity to review the methodology and results.

Used: News Factory APP - news discovery and automation - ChatGPT for Business