Milestones
- GenCast diffusion model beats ECMWF ensemble on 97% of tests4 Dec 2024Complete.
- WeatherNext 2 released, up to 8x faster, 15-day range17 Nov 2025Complete.
- WeatherNext 3 launched with hourly satellite-fed forecasts3 Sep 2026Complete.
Give sharper, more frequent and more accurate weather forecasts, especially rain, at a fraction of the computing cost of traditional physics simulations, including in regions that forecasts have long served poorly.
WeatherNext 3 launched Sep 2026: hourly global forecasts, now behind weather in Search, Gemini and Maps.
Updated 3 Sep 2026Checked 11 Oct0 updates this week
Instead of waiting for a physics model's analysis of current weather, it reads live geostationary satellite images and scattered station readings.
A generative network adds controlled randomness, so one model can produce many plausible forecasts; how far they spread tells forecasters how sure to be.
One model outputs gridded maps, tropical cyclone tracks and forecasts at individual stations, plus 100 m wind and solar data for clean-energy planning.
Running a trained network is cheap: WeatherNext 2 made a forecast in under a minute on Google's TPU chips, versus hours for physics-based models.
| Spec | WeatherNext 3 | WeatherNext 2 | Change |
|---|---|---|---|
| Forecast refresh | Every hourR (reported) | Every 6 hours | ▼ 5 h |
| Surface grid (temperature, moisture) | 5 kmR (reported) | 25 km | ▼ 20 km |
| Atmosphere grid (wind) | 25 kmR (reported) | — | —no comparable change |
| Rain forecasts, 1+ day ahead | Up to 50% more accurateR (reported) | — | —no comparable change |
| Lead time (WeatherNext 2) | Up to 15 daysR (reported) | — | —no comparable change |
R reported by the company
Google DeepMind
Gemini, AlphaFold, Genie, Veo, TPU chips
Google DeepMind is Alphabet's AI lab, formed in 2023 from DeepMind and Google Brain. It builds the Gemini models and leads in AI for science. AlphaFold won a share of the 2024 Nobel Prize in Chemistry.