DeepMind open-sources WeatherNext 2 and WeatherNext Cyclones with a Nature paper showing ~1 extra day of hurricane warning

Your AI and this story

Key facts

• Three-day WeatherNext Cyclones forecast about as accurate as prior systems at two days: ‘>24 hours lead time advantage’
• Released: WeatherNext Cyclones, WeatherNext 2, WeatherNext 2-mini (runs on a single TPU / free Colab)
• Licences: Apache 2.0 for code/notebooks, CC BY 4.0 for other materials — first DeepMind weather weights allowing commercial use
• Paper in Nature (s41586-026-10953-2)
• Partners: US National Hurricane Center, CIRA, UK Met Office; helped NHC forecast Hurricane Melissa’s 2025 rapid intensification
• Google blog Q&A (Sept 1) with DeepMind’s Ferran Alet: WeatherNext was trained on 50 years of historical weather data and runs on a single TPU; high-confidence AI forecasts helped the US National Hurricane Center warn of Hurricane Melissa’s rapid intensification from Category 1 to Category 5
• 2026 season (Local 10 Miami, Sept 10): DeepMind’s model was the most accurate hurricane track model of 2026 through early September in both the Atlantic and Pacific, beating NHC’s official track forecasts by up to ~30% at five days, the second year in a row it topped NHC at most lead times
• Hurricane Isaias (Oct 7–10, 2026; the latest first Atlantic hurricane of a season on record): AI models, including WeatherNext Cyclones, which NHC uses operationally, favoured an eastern landfall on the western Florida Panhandle (Cat 1, early Sat Oct 10), while physics-based models favoured Mississippi/Louisiana; NHC’s official track split the difference (Gizmodo, Oct 8). CNN (Andrew Freedman, Oct 8): NHC shaped its forecasts to closely match Google DeepMind’s AI model (named in the syndicated text as WeatherNext 3) rather than the European and GFS physics models, forecasting Category 2 from the first advisory; NHC director Michael Brennan: ‘The AI models have shown more consistency from one model run to the next’