Google DeepMind Launches WeatherNext 2 — AI Model Delivers Ultra-Fast, High-Resolution Forecasts

Google DeepMind, in collaboration with Google Research, has launched WeatherNext 2, its most advanced AI-based weather forecasting model to date.

The system promises faster, more accurate, and higher-resolution global predictions, according to a blog post by Google.

Powered by a novel architecture called a Functional Generative Network (FGN), WeatherNext 2 can generate hundreds of possible weather scenarios from a single input in under a minute using just one TPU. This represents a major speed boost — the model is eight times faster than its predecessor.

"The weather affects important decisions and high quality weather forecasting has a significant impact on people, businesses and society. In recent years AI-based modeling has dramatically enhanced what’s possible with weather forecasting and the new WeatherNext 2 models are more accurate, of higher resolution, and more efficient," Yossi Matias, Vice President, Head of Google Research, said.

In addition to performance gains, the model outperforms WeatherNext 1 on 99.9% of variables, such as temperature, humidity, and wind, across forecast lead times up to 15 days.

WeatherNext 2 also offers finer temporal resolution, delivering hourly predictions that are more useful for decision-making.

Google is rolling out WeatherNext 2 data through several platforms: it’s now accessible in Earth Engine, BigQuery, and via an early-access program on Vertex AI.

The model’s forecasts are already integrated into consumer-facing tools such as Google Search, Gemini, Pixel Weather, and the Maps Platform’s weather API — with more Google Maps rollout expected soon.

"WeatherNext 2 represents a significant step in translating breakthrough AI from the lab into high-impact, operational applications. We are making powerful tools available to the global community to accelerate scientific discovery and build a more resilient future," Matias added.