Research
1 / 5
Three themes dominate 2024–2026 arXiv AI weather research: large foundation models, physics-constrained architectures, and probabilistic ensembles.
2 / 5
Microsoft Aurora, Aardvark Weather, and NeuralGCM deliver strong benchmarks but lack native high-resolution, high-frequency, or energy-specific outputs.
3 / 5
Jua’s EPT-2 and EPT-1.5 beat Aurora, GraphCast, and ECMWF HRES on energy-critical variables while training faster and supporting native any-Δt forecasting.
4 / 5
EPT-2e’s 10-member ensemble outperforms the 50-member ECMWF ENS on RMSE and CRPS at nearly every lead time, with four daily updates and far lower compute cost.
5 / 5
See EPT-2 benchmarked live against your current forecast provider in under five minutes.
Explore top arXiv AI weather papers on foundation models, physics constraints & ensembles. See how Jua's EPT-2 outperforms Aurora, GraphCast & ECMWF.