Google DeepMind open sources WeatherNext 2 and WeatherNext cyclones
Drafted by our automated systems from the sources listed at the end of this article, and reviewed before publication. Every fact here should be traceable to one of those links; we publish no benchmark, price or figure that a linked source does not state. Where a company makes a claim about its own product, we report it as their claim.
Google DeepMind announced on 6 August 2026 that it is open sourcing the code and model weights for WeatherNext 2 and WeatherNext Cyclones. According to Google DeepMind, the technology gives forecasters an extra day of warning on average, noting that its three-day forecasts match what previous models achieved for two days. The company states that this corresponds to roughly a decade of meteorological progress. The release includes the WeatherNext Cyclones models that operated during the hurricane season and WeatherNext 2, an update operationalised in October.
Training data and model architecture
Google DeepMind reports that the WeatherNext model was co-trained on nearly 20 terabytes of global atmospheric data alongside the historical IBTrACS database, which covers nearly 5,000 historical storms. The company states that the model uses Functional Generative Networks to produce ensembles of different predictions. According to Google DeepMind, the system can generate a single 15-day forecast in less than a minute on a TPU. The post states that WeatherNext Cyclones operates at a data resolution of 28x28km, which is described as 100 times coarser than traditional models. A smaller variant named WeatherNext 2-mini operates at a 111x111km resolution.
Real-world testing and future questions
Google DeepMind reports that its research has already had real-world impact. During the 2025 hurricane season, the model helped the National Hurricane Center make a forecast for Hurricane Melissa by predicting rapid intensification and landfall in Jamaica. The company states that it now predicts 1,000 possible scenarios for each cyclone. Google DeepMind notes that it remains an open research question to fully understand how the models produce such accurate predictions at coarser resolutions, and the organisation hopes the wider research community will help investigate this.
Sources
- WeatherNext: AI model achieves breakthrough in forecasting cyclones — Google DeepMind, 6 August 2026