{"id":"weathernext-3","slug":"weathernext-3","name":"WeatherNext 3","orgId":"google","family":"weathernext-3","releaseDate":"2026-09-03","availability":"ga","openWeights":false,"license":null,"paramsTotal":380000000000,"paramsSource":"epoch-estimate","contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/","hfRepo":null,"modalities":["weather"],"aliases":["WeatherNext 3","weathernext-3","WeatherNext3","weather next 3","Google WeatherNext 3","Google DeepMind WeatherNext 3"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-11-12","announced":"2026-09-03","blurb":"Google DeepMind and Google Research's global weather forecasting model, announced September 3, 2026. Official pages say it ingests live geostationary satellite data to refresh forecasts every hour, at up to 5 km for station-trained surface variables. It powers weather experiences in Google Search, the Gemini app, and Google Maps, and is used in the Google Maps Platform Weather API. Raw forecast datasets on Cloud Storage, Earth Engine, and BigQuery require an access request. Official docs state that WeatherNext 3 is not open source.","api":"live","openWeights":"no","facts":[{"label":"Type","value":"Global AI weather forecasting model from Google DeepMind and Google Research. Official docs describe a Functional Generative Network (FGN) mesh transformer with 64 ensemble members, hourly initialization, and a 15-day horizon on the 6-hourly cycles (Google blog, developers model page, and paper)"},{"label":"Resolution","value":"Official blog and docs: station-trained surface variables such as temperature and moisture at 5 kilometers (0.05°), other surface variables at 10 kilometers (0.1°), and atmospheric variables such as wind at 25 kilometers (0.25°). WeatherNext 2 produced forecasts on a 25-kilometer grid in 6-hour increments (Google blog and developers model page)"},{"label":"Hosted access","value":"Official blog: rolling out in Google Search, the Gemini app, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine as of September 3, 2026. The Maps Platform Weather API is a documented self-serve API (API key / Maps Platform account). WeatherNext 3 forecast datasets on Google Cloud Storage, Earth Engine, and BigQuery are available after submitting a data-access form; official docs say requests are typically approved within 5-7 business days (Google blog, Maps Platform Weather API docs, and WeatherNext access guide)"},{"label":"Weights","value":"Official docs: WeatherNext 3 is available as an operational forecast service but is not open source. Earlier WeatherNext 2, Gen, and Graph research models have public code and pretrained weights on GitHub; those are separate models (Google WeatherNext open-source guide)"},{"label":"License","value":"No public weight license for WeatherNext 3. Real-time forecast data is under the GDM Real-Time Weather Forecasting Experimental Data Terms of Use; historical data (one hour old or more) is licensed under CC BY 4.0. Consumer and Maps Platform use is under the applicable Google and Maps Platform terms (Google WeatherNext access guide)"},{"label":"Related","value":"Successor to WeatherNext 2. Official docs also list earlier research models WeatherNext Gen and WeatherNext Graph. No WeatherNext family hub is on isitsota yet"}],"claims":["Lab claim: self-reported as providing a global weather picture roughly five times sharper than WeatherNext 2 (Google blog).","Lab claim: self-reported Continuous Ranked Probability Score (CRPS) improvements of up to 60% against IMERG, 30% for MRMS, and 10% against rain gauge measurements for early lead times in medium-range global forecasts, plus up to 50% more accurate precipitation forecasts when planning a day or more ahead (Google blog).","Lab claim: self-reported up to a 50% reduction in Brier score and CRPS compared with numerical weather prediction baselines when evaluated against IMERG observations (Google WeatherNext model documentation)."],"sources":[{"kind":"lab","title":"Introducing WeatherNext 3 (Google blog)","url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/","date":"2026-09-03"},{"kind":"lab","title":"WeatherNext 3 (Google DeepMind)","url":"https://deepmind.google/science/weathernext/"},{"kind":"lab","title":"WeatherNext 3 model documentation","url":"https://developers.google.com/weathernext/guides/models"},{"kind":"lab","title":"Accessing WeatherNext forecasts","url":"https://developers.google.com/weathernext/guides/access-forecast"},{"kind":"lab","title":"WeatherNext open source models","url":"https://developers.google.com/weathernext/guides/osmodel"},{"kind":"lab","title":"Google Maps Platform Weather API","url":"https://mapsplatform.google.com/maps-products/weather/"},{"kind":"lab","title":"WeatherNext 3 paper (arXiv:2609.03582)","url":"https://arxiv.org/abs/2609.03582","date":"2026-09-03"}],"firstBenchmark":null},"organization":{"id":"google","name":"Google DeepMind","slug":"google","country":null,"url":"https://deepmind.google","color":"#4285F4","mark":"G"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Sep 3, 2026; no independent results yet.","lead":"As of Nov 22, 2026, WeatherNext 3 is pending benchmarks on isitsota: announced Sep 3, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}