WeatherNext 3 Guide: Specs, Forecasts, Access and Limits
WeatherNext 3 is Google DeepMind's hourly global weather model. It combines live satellite input with a 64-member ensemble and produces forecasts out to 15 days at up to roughly 5 km for station-targeted temperature and humidity. This independent guide is the fastest route to the right WeatherNext 3 page, whether you want a local forecast, a model explanation, API access or licensing terms. Published October 20, 2026 and reviewed by the WeatherNext 3 editorial team.
7-day free trial, then $19/mo · cancel anytime · compare plans
What should you know first?
WeatherNext 3 changes the useful unit of a forecast from a daily summary to an hourly sequence. Read the timing, location and uncertainty together.
Hourly refresh
The model is initialized every hour with live geostationary satellite observations, so an intraday plan can be checked against a newer global run.
Multiple resolutions
Station-targeted temperature and humidity reach roughly 5 km, surface grids use about 10 km and pressure-level fields around 25 km.
Probabilistic output
The 64-member ensemble describes a range of plausible futures. The mean is one summary, not the whole forecast.
Fifteen-day horizon
The longest operational horizon reaches 15 days, with confidence and decision value changing substantially across that range.
WeatherNext 3 at a glance
Four facts explain most of what makes the model different from a conventional daily weather page.
1 hour Designed to initialize a new global forecast every hour of the day.
Designed to initialize a new global forecast every hour of the day.
64 Ensemble members describe a distribution of possible atmospheric outcomes.
Ensemble members describe a distribution of possible atmospheric outcomes.
15 days Maximum operational forecast horizon for the flagship WeatherNext 3 product.
Maximum operational forecast horizon for the flagship WeatherNext 3 product.
5 km Station-targeted 2 m temperature and humidity are produced at up to 0.05 degrees.
Station-targeted 2 m temperature and humidity are produced at up to 0.05 degrees.
Start with the question you have
These pages explain the model, the product experience and the core workflow without assuming technical weather knowledge.
What is WeatherNext 3?
A plain-language introduction to the hourly AI model, its variables and what changed from earlier forecasts.
How to use WeatherNext 3
Move from a place and a decision to a threshold, a time window and a usable forecast.
WeatherNext 3 forecast
How to read the hourly sequence, variables and resolution without over-interpreting one frame.
WeatherNext 3 15-day forecast
What the long-range horizon can answer, how confidence changes and when to use the ensemble instead of the mean.
Weather variables and operational guides
Use the variable page that matches the decision. Each guide separates what the model can resolve from what still requires local data or official warnings.
Precipitation forecast
How rain and snow are modelled, how to read hourly arrival windows and what the benchmark gains mean.
Temperature forecast
Read the hourly temperature curve, humidity and local terrain effects at roughly 5 km.
Wind forecast
Understand 100 m wind, direction, hourly windows and the limits of a 10 km surface grid.
Solar forecast
Use solar radiation and cloud cover for photovoltaic planning without confusing irradiance with guaranteed output.
Radar and rain layers
Why a forecast rain layer is not live radar and how to pair observations with the next hourly window.
Hurricane forecast
Track hurricane windows with hourly fields while keeping official warnings authoritative.
Typhoon forecast
Apply the hourly forecast to western Pacific, South China Sea and Indian Ocean cyclone windows.
Model, research and comparisons
Go deeper into the architecture, paper, benchmark claims and the models that came before WeatherNext 3.
WeatherNext 3 model
Architecture, resolution, satellite ingestion and the shift from six-hourly to hourly initialization.
WeatherNext 3 accuracy
What the reported improvements measure, which baselines they use and how to test a local decision.
WeatherNext 3 paper
The key claims of arXiv 2609.03582 without losing the assumptions and caveats.
WeatherNext 3 DeepMind
The research team, the WeatherNext family and the sources worth reading directly.
Weather Lab
How Google's research interface exposes global forecast layers and where a research view differs from a product.
vs WeatherNext 2
Resolution, cadence, input data and decision windows across the two model generations.
vs GraphCast
GraphCast, GenCast and WeatherNext 3 compared on cadence, resolution and ensembles.
vs ECMWF
A fair comparison with ECMWF products across resolution, operational maturity and forecast objectives.
App, API, data and licensing
Choose the access path that matches the job: a forecast on a phone, a developer API, research data or a commercial review.
WeatherNext 3 app
There is no standalone WeatherNext 3 app. See how the browser experience works across devices.
Google WeatherNext 3 app
Where the model appears inside Google's own apps and why unofficial APK listings deserve suspicion.
iOS app route
The practical route onto iPhone without an App Store listing.
WeatherNext 3 API
How to request hourly forecast fields programmatically and what to check before production use.
Data access
Compare BigQuery, Earth Engine and Cloud Storage, including access and operational tradeoffs.
API pricing
Understand the metered Google Cloud paths and when a managed API is the cheaper route.
WeatherNext 3 cost
What is free in Google products, what cloud usage costs and what a subscription includes here.
WeatherNext 3 license
Separate code, weights, historical data and real-time forecasts before planning a public product.
Choose a path in four steps
Start with the job, not with a keyword. The category tells you where the useful answer sits.
Define the decision
A commute, a wind-farm window, a solar dispatch plan and an API integration need different variables and different levels of uncertainty.
Pick the variable
Start with rain, temperature, wind or radiation, then add humidity, cloud cover or another field only when it changes the decision.
Choose the horizon
Use hourly detail for today and the next 48 hours, then shift toward ensemble ranges and scenario planning as lead time grows.
Choose the access path
Use the browser for a forecast, the API for a product and the research datasets when a team needs raw fields or historical analysis.
WeatherNext 3 guide questions
Direct answers to the questions that usually send people across several WeatherNext 3 pages.
Next: wind forecasting, solar forecasting or licensing.
Explore the forecast behind the guide
Choose a location, inspect the hourly sequence and compare variables in one workspace. Seven-day free trial, cancel anytime. Related: reading the forecast and the 15-day horizon.