Robots Atlas>ROBOTS ATLAS
WeatherNext 2

WeatherNext 2

2ย ยทย Family: WeatherNext
Google DeepMind and Google Research's latest weather-forecasting model โ€” generative, ensemble; about 8x faster than its predecessor and better at extreme events.
โœ“ Activeโœ“ Public accessScientific AISpecialized AI๐Ÿ“ WeatherNext
Parameters
Nieujawnione publicznie
parameters
Access:APIHostedDeployment:โ˜ Cloud

Overview

WeatherNext 2 is the latest model in the WeatherNext family โ€” developed jointly by Google DeepMind and Google Research โ€” for AI-based weather forecasting. The model generates fast, probabilistic (ensemble) forecasts of the atmospheric state, including wind speed and direction, precipitation and pressure.

Compared to the previous generation, WeatherNext 2 is about 8x faster (a forecast can be generated in under a minute on a single TPU accelerator) and predicts extreme events more accurately. The extra speed allows analyzing many more possible scenarios per forecast. The model produces forecasts in six-hour steps, four times a day; in the experimental Weather Lab it supports cyclone prediction up to 15 days in advance.

WeatherNext 2 is available through Google Cloud (Vertex AI / Gemini Enterprise Agent Platform), BigQuery, Google Earth Engine and the Google Maps Platform Weather API, and powers Google products: Search, Gemini and Pixel Weather. Note: the architecture details, exact spatial resolution and official release date are not stated on the accessible Google pages.

Classification
Scientific AISpecialized AI
Family: WeatherNext
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
๐Ÿงฉ Parameters: Nieujawnione publicznie
๐Ÿ“ฅ Input: structured data, time series

Technical specification

Parameters
Nieujawnione publicznie
parameters
License
proprietary
Hardware requirements
Available only through Google cloud infrastructure (Vertex AI, BigQuery, Earth Engine, Weather API); inference on TPU accelerators.
Modalities
โฌ‡ Input
structured_datatime_series
โฌ† Output
structured_dataanalytical_reports

Capabilities and applications

Native model capabilities
Structured output
Producing data in structured formats such as JSON.
Category: structured_generation