Google has open-sourced its WeatherNext 2 and WeatherNext Cyclones AI models to improve cyclone forecasting accuracy. The models achieve state-of-the-art results in predicting track, intensity, and wind structure, providing forecasters with an average of one extra day of predictive accuracy compared to current methods.

  • Google's WeatherNext models predicted Hurricane Melissa's rapid intensification and landfall in Jamaica during the 2025 season.
  • Meta researchers investigated how modalities learn together during pretraining, identifying architectures that promote cross-modal synergy.
  • Meta's approach achieved strong generation performance using only 5% of the compute budget.

The open-sourced weather models aim to empower the research community and help build more resilient communities through advanced warnings. Meanwhile, Meta's findings on cross-modal learning offer efficient recipes for training multimodal systems.