How is WeatherNext 3 different from WeatherNext 2?

WeatherNext 2 produced forecasts on a 25-kilometer grid in 6-hour increments. WeatherNext 3 forecasts hourly and at 5-kilometer resolution for surface variables, which Google describes as a global weather picture roughly five times sharper. WeatherNext 3 also continuously ingests live geostationary satellite data, which WeatherNext 2's release did not describe.

How much more accurate is WeatherNext 3 at forecasting rain?

Google reports that WeatherNext 3 improves the Continuous Ranked Probability Score for medium-range precipitation by up to 60% against NASA's IMERG satellite data, 30% against MRMS and 10% against rain gauge measurements at early lead times. For longer-range planning, Google says precipitation forecasts are up to 50% more accurate, with the biggest gains in historically unreliable regions.

Where can developers access WeatherNext 3 forecasts?

WeatherNext 3 forecasts are available to developers through Google Earth Engine and BigQuery, with bulk downloads from Google Cloud Storage and programmatic access via the Google Maps Platform Weather API. Google documents the options at developers.google.com/weathernext. The same forecasts also power consumer surfaces including Google Search, Gemini and Google Maps.

What data does WeatherNext 3 forecast?

WeatherNext 3 covers surface variables such as temperature at 5-kilometer resolution, other surface fields at 10 kilometers, and atmospheric variables at 25 kilometers, all on an hourly step. Precipitation is the headline improvement. Google validates the model against NASA's IMERG satellite product, the MRMS radar-based system and ground rain gauges.

Google DeepMind · 2026-09-03 · major

WeatherNext 3 — DeepMind's weather model goes hourly at 5km

WeatherNext 3 is Google DeepMind's new global weather model. It produces hourly forecasts at 5km resolution, about five times sharper than WeatherNext 2, and improves precipitation scores by up to 60% against NASA's IMERG data.

WeatherNext 3 title graphic from Google DeepMind

DeepMind's global forecast model moves from 6-hour, 25km predictions to hourly ones at 5km.

Key specs

Sharper than weather next 2	5x
Precipitation crps vs imerg	up to 60% better

Quick facts

Maker	Google DeepMind and Google Research
Released	September 3, 2026
Forecast step	Hourly
Resolution	5km surface variables, 10km other surface fields, 25km atmospheric
Previous model	WeatherNext 2 — 25km grid, 6-hour increments
Live inputs	Geostationary satellite data, ingested continuously
Availability	Google Search, Gemini, Maps, Maps Platform Weather API, Earth Engine, BigQuery, Cloud Storage

What is it?

WeatherNext 3 forecasts the whole planet hour by hour rather than in six-hour jumps, at 5-kilometer resolution for surface variables such as temperature. Google DeepMind and Google Research call it their most advanced and accurate global weather model to date. Other surface fields run at 10 kilometers and atmospheric variables at 25 kilometers.

How does it work?

The extra detail comes from running on a much finer grid than WeatherNext 2, which produced forecasts on a 25-kilometer grid in 6-hour increments — the new model gives a global picture roughly five times sharper. WeatherNext 3 also takes in live geostationary satellite data as it runs, which keeps its starting state current and matters most for short-range rain.

Why does it matter?

Rain is where WeatherNext 3 gains most. For medium-range precipitation Google reports Continuous Ranked Probability Score improvements of up to 60% against NASA's IMERG satellite data, 30% against MRMS and 10% against rain gauges at early lead times, and up to 50% more accurate precipitation forecasts for longer-range planning, with the largest gains in regions where forecasts have historically been least reliable. The output already feeds Google Search, Gemini and Maps, and developers can pull it from Earth Engine, BigQuery and the Maps Platform Weather API.

Who is it for?

developers and researchers building on weather data

Frequently asked questions

How is WeatherNext 3 different from WeatherNext 2?: WeatherNext 2 produced forecasts on a 25-kilometer grid in 6-hour increments. WeatherNext 3 forecasts hourly and at 5-kilometer resolution for surface variables, which Google describes as a global weather picture roughly five times sharper. WeatherNext 3 also continuously ingests live geostationary satellite data, which WeatherNext 2's release did not describe.
How much more accurate is WeatherNext 3 at forecasting rain?: Google reports that WeatherNext 3 improves the Continuous Ranked Probability Score for medium-range precipitation by up to 60% against NASA's IMERG satellite data, 30% against MRMS and 10% against rain gauge measurements at early lead times. For longer-range planning, Google says precipitation forecasts are up to 50% more accurate, with the biggest gains in historically unreliable regions.
Where can developers access WeatherNext 3 forecasts?: WeatherNext 3 forecasts are available to developers through Google Earth Engine and BigQuery, with bulk downloads from Google Cloud Storage and programmatic access via the Google Maps Platform Weather API. Google documents the options at developers.google.com/weathernext. The same forecasts also power consumer surfaces including Google Search, Gemini and Google Maps.
What data does WeatherNext 3 forecast?: WeatherNext 3 covers surface variables such as temperature at 5-kilometer resolution, other surface fields at 10 kilometers, and atmospheric variables at 25 kilometers, all on an hourly step. Precipitation is the headline improvement. Google validates the model against NASA's IMERG satellite product, the MRMS radar-based system and ground rain gauges.

Try it

https://developers.google.com/weathernext