WeatherNext 3
WeatherNext 3 is Google's September 2026 AI weather-forecasting model. It is live globally in Google Search, Gemini, Maps, Maps Platform, and Google Cloud, while developers and researchers can access forecast data through BigQuery, Earth Engine, and Cloud Storage.
WeatherNext 3 moves a specialist forecasting model into both consumer products and operational data surfaces. Its hourly runs, multiple spatial resolutions, forecast horizons, ensemble output, migration changes, and safety boundary matter more to a user than treating it like a general-purpose LLM API.
Google's launch post, DeepMind overview, and developer guides confirm the September 3 launch, global integrations, live satellite mosaics, hourly initialization, forecast horizons, resolutions, data-access routes, migration notes, and the requirement not to treat the experimental data as an official warning or sole decision source. Performance figures remain Google-reported unless separately identified.
- Read current AI-assisted weather forecasts through Google consumer products.
- Analyze global forecast fields in BigQuery or Earth Engine.
- Build renewable-energy, logistics, agriculture, insurance, or planning workflows with hourly forecast refreshes.
- Compare probabilistic ensemble outcomes instead of relying on one deterministic forecast.
- Evaluate a WeatherNext 2 migration against current variables, grids, cadence, and warning requirements.
The model combines recent observations with live geostationary satellite mosaics and initializes every hour. Google publishes station outputs at roughly 5 km, gridded surface variables at roughly 10 km, and pressure-level variables at roughly 25 km. Six-hourly cycles extend to 15 days; interim hourly cycles extend to 48 hours.
- Hourly initialization supports fresher operational decisions than a small set of daily forecast cycles.
- A 64-member ensemble represents multiple plausible weather outcomes instead of one deterministic path.
- Available variables, cadence, horizon, and resolution differ by dataset and product surface.
Google says WeatherNext 3 is used in Search, Gemini, Maps, Maps Platform, and Google Cloud. Developers and researchers consume its forecast data through BigQuery, Earth Engine, or Cloud Storage rather than selecting a public LLM-style request model ID. That is why GetLLMs tracks the model as an entity without inventing a `/models` API record.
WeatherNext 3 adds hourly precipitation, renewable-energy variables, and new satellite-driven products, while some variable names and dataset layouts differ from WeatherNext 2. A production consumer should compare the active guide, schema, units, spatial grid, timestamp semantics, missing-data behavior, and retrieval cost before switching.
Google reports improved forecast quality, including precipitation gains and better handling of rapidly developing weather, and links technical evaluation material. Those are publisher-reported results under specified datasets and metrics. Validate the locations, lead times, variables, baselines, and operational loss function that matter to the actual decision.
Google labels the developer data experimental, provided as-is, and unsuitable as the sole basis for safety-critical decisions. WeatherNext 3 should complement official meteorological agencies, local alerts, domain expertise, and redundant observations rather than replace them.
WeatherNext 3 FAQ
Common questions about WeatherNext 3.
Is WeatherNext 3 available now?+
Yes. Google says WeatherNext 3 is live globally across Search, Gemini, Maps, Maps Platform, and Google Cloud. Exact features and datasets still differ by product, region, account, and access surface.
How can developers access WeatherNext 3?+
Google publishes WeatherNext 3 forecast data through BigQuery, Earth Engine, and Cloud Storage. The reviewed developer surface does not provide a public LLM-style model ID, so integration starts with the relevant dataset and schema rather than a chat-completions request.
What resolution and forecast horizon does WeatherNext 3 provide?+
Google documents roughly 5 km station, 10 km gridded-surface, and 25 km pressure-level products. Six-hourly cycles extend to 15 days, while interim hourly cycles extend to 48 hours. Check the selected dataset because variables and availability differ.
Can WeatherNext 3 replace official weather warnings?+
No. Google says the experimental forecast data is provided as-is and should not be the sole source for safety-critical decisions. Use official meteorological alerts and local authorities for warnings and emergency action.
What changes when migrating from WeatherNext 2?+
WeatherNext 3 changes some variable names and dataset organization while adding hourly initialization, hourly precipitation, renewable-energy variables, and satellite-driven products. Compare the current migration guide, schema, units, grid, timestamps, and downstream assumptions before switching.