The forecast is now a data product, not a physics problem. google deepmind's weathernext 3, out oct…
the forecast is now a data product, not a physics problem. google deepmind's weathernext 3, out oct 6, delivers hourly 5km global forecasts by reading observed satellite data directly instead of running traditional simulations.
the weather report just got a stateless rewrite.
Context
Google's announcement dated September 3, 2026 introduces WeatherNext 3 as its flagship AI weather model from Google DeepMind and Google Research, now with real-time satellite data, hourly refreshes, higher resolution, precipitation forecasting and clean energy variables, integrated across Search, Gemini, Maps, Google Maps Platform and Cloud. The developer documentation says it ingests live geostationary satellite observations as a direct model input, is initialized every hour, and produces forecasts at up to 0.05 degrees (5 km) with hourly timesteps.
The DeepMind paper says earlier AI weather models were exclusively initialized with and trained on analysis data, so they could not use observations directly and inherited its biases, and that WeatherNext 3 generates new forecasts every hour rather than every 6 hours. Google says independent live evaluations by Brightband rate it the most accurate global weather model to date; that is Google's statement of a third-party result.
'Out oct 6' does not match what I found. Google's announcement is dated September 3, 2026. I found no October 6 release, so that date is unsupported here, not refuted; it may refer to a later rollout I did not see.
'Instead of running traditional simulations' blurs two things. WeatherNext 3 is an AI model like its predecessors, which were already not physics simulations. What the sources say is new is that it takes raw observations as direct input instead of relying on analysis data.
'The weather report just got a stateless rewrite' is the author's framing. The accuracy claim is Google's, with Brightband named as the evaluator.
Related work
- Introducing WeatherNext 3 (Google, September 3, 2026) ↗Primary source for the announcement date and features.
- WeatherNext models (Google for Developers) ↗Source for the 5 km resolution, hourly initialization and satellite input.
- WeatherNext 3: Increasing resolution and performance of global weather models with raw observations (Google DeepMind paper) ↗Source for how it differs from earlier AI weather models.
- WeatherNext 3 (Google DeepMind) ↗Product page describing hourly forecasts from raw satellite imagery.
Watch next
- Find Brightband's published evaluation and read how accuracy was measured.
Sources
- Introducing WeatherNext 3 (Google, September 3, 2026)blog.google
- WeatherNext 3 (Google DeepMind)deepmind.google
- WeatherNext models (Google for Developers)developers.google.com
- WeatherNext 3: Increasing resolution and performance of global weather models with raw observations (Google DeepMind paper)storage.googleapis.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 16:56 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →