- Google has launched a new AI-powered forecasting model called WeatherNext 2
- WeatherNext 2 has faster, more accurate, and higher-resolution predictions by simulating hundreds of possible weather outcomes in under a minute
- WeatherNext 2 is now powering forecasts in Google Search, Gemini, Pixel, and Maps
Google is overhauling your weather forecast with AI that thinks in probabilities. Rather than new radar towers or satellite launches, the new WeatherNext 2 AI-based forecasting model developed by Google DeepMind and Google Research delivers results up to eight times faster than traditional systems and can predict hundreds of possible weather outcomes from a single starting point at higher resolution and with better accuracy than its predecessors.
WeatherNext 2 is being integrated into many of the most popular Google platforms, including Google Search, Gemini, Pixel Weather, and Maps, with a broader rollout coming soon via the Google Maps Platform Weather API.
What makes this upgrade more than just a back-end refresh is the sheer scale of its ambition. WeatherNext 2 is built to account for uncertainty in unusual ways. Where older models might spit out a single most-likely outcome, WeatherNext 2 can generate hundreds of potential futures, allowing forecasters and you to see a full spread of possibilities.
It also means your forecast might not just say “Rain, 40% chance,” but instead show multiple coherent outcomes for your afternoon walk, with better insight into what could actually happen, and when.
Watch On
WeatherNext 2 uses what Google calls a Functional Generative Network (FGN). The model doesn’t rely solely on finished forecasts or entire weather systems; instead, it’s trained on individual, standalone variables such as temperature, wind speed, and humidity. The model then works out how those variables interact to create “joints,” the complex, real-world patterns like storm fronts, heat waves, or regional wind shifts.
Google claims this architecture enables WeatherNext 2 to outperform even its own previous best-in-class model, delivering more accurate forecasts for 99.9% of variables across for up to 15 days out.
It’s also much faster at making its predictions, completing a full forecast in under a minute. By comparison, traditional physics-based forecasting can take hours on a supercomputer. This efficiency allows for more frequent, more detailed forecast updates.
Weathering the future
After considerable testing, Google Gemini will start showing forecasts based on WeatherNext 2’s outputs, as will Google Maps. The average person could theoretically gain many benefits from the upgrade. Weather forecasting is one of those invisible systems that underpins an extraordinary range of decisions. Make it more accurate, and you remove a thousand tiny sources of stress from people’s days.
There are bigger implications, too. For instance, with smarter weather predictions, renewable energy providers can better estimate wind and solar output, and emergency services can respond with more precision when forecasts capture uncertainty rather than masking it.
That emphasis on uncertainty is key. Forecasting isn’t about being perfectly right, but preparing wisely for what might happen. By providing a range of physically realistic, interconnected scenarios, WeatherNext 2 nudges forecasting toward something more strategic.
This may not solve the chaos of climate change and related natural disasters, but it might be a boon for those planning ways to better address them. AI-powered forecasting starts to look like essential infrastructure.
Better data means making better decisions. And for the weather, that means a lot more than just helping us decide what kind of coat we need in the morning.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course you can also follow TechRadar on TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
The best business laptops for all budgets
