Google's WeatherNext 2 delivers 8x faster, more accurate AI weather forecasts, with 15-day outlooks and frequent updates across Maps, Search, and Gemini.
© B. Naumkin
Google has unveiled WeatherNext 2, an upgraded AI-driven weather forecasting model built to deliver markedly better accuracy and faster updates. The system will underpin weather information across Pixel Weather, Google Search, the Gemini app, and Google Maps. The move points to a concerted push to make AI the default for forecasts throughout its ecosystem.
According to the company, WeatherNext 2 produces forecasts eight times faster than earlier models and generates four updates every six hours, covering up to 15 days. It also outperforms WeatherNext 1 in accuracy across 99% of meteorological variables—temperature, precipitation, pressure, wind, and humidity. On paper, that’s a notable jump for a second-generation release.
The key change is a shift to a Functional Generative Network (FGN) architecture, replacing the GNN and diffusion models used in the first version. Leveraging Google’s TPU ASICs, WeatherNext 2 can generate a full forecast in under a minute; comparable physics-based models running on a supercomputer would take about an hour. That gap helps explain Google’s pivot toward data-driven methods in this domain.
Google plans to deploy WeatherNext 2 across all of its services that display weather. In the coming weeks, the new system will begin producing forecasts in Google Maps. Users of Pixel Weather, Gemini, and Google Search will also gradually see improved accuracy and more stable readings. The steady rollout suggests the company is prioritizing reliability as it scales.
WeatherNext 2 analyzes interdependent meteorological variables without solving physical equations on a supercomputer; instead, the AI looks for recurring patterns in massive datasets. As a result, a 15-day outlook can be produced in about a minute and, when needed, thousands of distinct scenarios can be generated.