DeepMind's WeatherNext 2 Outperforms Human Forecasters in Tropical Cyclone Predictions

Google DeepMind's WeatherNext 2 model has outperformed traditional methods in predicting tropical cyclones, surpassing even human forecasters' accuracy. This achievement is significant for the crypto ecosystem, which relies on accurate real-time weather data for prediction markets and decentralized finance (DeFi) insurance protocols.

The model generates probabilistic ensemble forecasts up to 15 days ahead with hourly resolution, running eight times faster than its predecessor models. During the 2025 Atlantic hurricane season, the National Hurricane Center tested WeatherNext 2, which outperformed conventional methods across key performance metrics for cyclone forecasting.

What's more intriguing is that researchers behind the model don't yet fully understand how it achieves this level of accuracy using lower-resolution weather data inputs. Earlier components of the WeatherNext series have been open-sourced, and Google DeepMind has indicated that WeatherNext itself will be made available to the public.