DeepMind unveils WeatherNext 2 for probabilistic forecasting at NeurIPS
The new model runs 8x faster than predecessors, predicting extreme weather events with higher accuracy.
A leap in meteorological AI
Google DeepMind presented its latest breakthrough, WeatherNext 2, at the NeurIPS 2025 conference on Wednesday. The model represents a significant advance in AI-driven meteorology, capable of running 8x faster than previous iterations while providing high-resolution forecasts down to the one-hour mark. Unlike traditional physics-based simulations that require massive supercomputing resources, WeatherNext 2 runs efficiently on Google's TPU hardware.
The model utilizes a "Functional Generative Network" (FGN) to generate thousands of possible weather scenarios from a single set of initial conditions in under a minute. This speed allows meteorologists to run "ensemble" forecasts at a …
Archive Access
This article is older than 24 hours. Create a free account to access our 7-day archive.