Google DeepMind Launches WeatherNext 2 for Faster Forecasts
Google DeepMind unveils WeatherNext 2, an AI model offering 8x faster forecasts with hour-by-hour accuracy up to 15 days ahead.
Google DeepMind Launches WeatherNext 2 for Faster Forecasts
Google DeepMind and Google Research have unveiled WeatherNext 2, a groundbreaking AI weather forecasting model that significantly advances meteorological prediction technology. The new system generates forecasts eight times faster than its predecessor, offering hour-by-hour predictions up to 15 days in advance.
Announced on November 17, 2025, WeatherNext 2 transforms the speed and reliability of weather prediction by utilizing advanced machine learning techniques to process complex atmospheric data. It surpasses Google's previous model on 99.9% of variables, including temperature, wind, humidity, and precipitation across all forecast lead times.
Technical Innovation
Functional Generative Network (FGN)
At the core of WeatherNext 2's capabilities is a novel AI modeling approach called a Functional Generative Network (FGN). This approach marks a departure from traditional physics-based forecasting methods by injecting "noise" into the model framework, allowing for multiple coherent weather scenarios while maintaining physical realism.
Computational Efficiency
WeatherNext 2's computational efficiency is a standout feature. It generates hundreds of possible weather outcomes from a single starting point in less than a minute on a single Tensor Processing Unit (TPU). This acceleration enables meteorologists to explore a wider range of possible scenarios, including low-probability but catastrophic events.
The model operates by generating four six-hour forecasts per day, refining accuracy through iterative processes. This capability is crucial for predicting large, interconnected atmospheric systems.
Scenario Prediction
WeatherNext 2's ability to predict hundreds of possible weather outcomes from a single input represents a paradigm shift. Instead of a single deterministic forecast, the model generates an ensemble of scenarios, capturing the full range of atmospheric possibilities. This is particularly valuable for extreme weather events, aiding in disaster preparation and response.
Real-World Applications
Google has integrated WeatherNext 2 into consumer-facing applications and enterprise platforms. It powers forecasts across Search, Gemini, Pixel Weather, and Google Maps Platform's Weather API, with further integration into Google Maps forthcoming.
Enterprise and Research Access
WeatherNext 2 is available to businesses, scientists, and developers through Google Cloud's Vertex AI platform, BigQuery, and Google Earth Engine. This democratization enables customized models for various sectors, including agriculture, energy, and disaster management.
Cyclone Prediction
Earlier in 2025, Google DeepMind demonstrated WeatherNext's impact through experimental cyclone predictions. The model predicts a cyclone's path, intensity, structure, and size up to 15 days in advance, providing invaluable information for coastal communities and emergency management.
Performance and Competitive Advantage
WeatherNext 2 maintains global coverage with training data from 2022 to the present. Its 15-day forecast horizon extends the original model's capability, offering decision-makers additional predictive insight.
The model's advancements through the FGN architecture and enhanced graph transformer framework make it the recommended starting point for new projects requiring state-of-the-art forecasting.
Implications for Climate Response
AI-driven weather forecasting like WeatherNext 2 is crucial for responding to climate change and extreme weather events. By providing faster, comprehensive scenario analysis, it supports better decision-making across various sectors.
Future Developments
Google DeepMind plans to expand WeatherNext 2's capabilities by integrating new data sources and broadening access. Its availability across multiple platforms positions it as a foundational tool in global weather forecasting infrastructure.
The introduction of WeatherNext 2 marks a transformative moment in meteorological technology, combining AI-accelerated computation with rigorous atmospheric science to deliver faster, more accurate weather predictions.
