In partnership with
Google DeepMind and Google Research have introduced WeatherNext 3, an AI system for global weather forecasting.
The announcement was published on September 3, 2026, with a matching research paper on arXiv.
The model differs from earlier systems by using recent satellite observations directly in its forecast process.
It combines low-latency geostationary satellite mosaics with traditional atmospheric analysis data.
How can AI power your income?
Ready to transform artificial intelligence from a buzzword into your personal revenue generator
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
The research paper says the model can produce new forecasts every hour.
Many global forecasting systems have traditionally worked with analysis fields produced at longer intervals.
The paper says those analysis fields can be several hours old by the time they reach a forecasting system.
That delay matters most for rapidly changing variables such as rainfall and surface temperature.
WeatherNext 3 produces outputs at several spatial scales.
Google says temperature and moisture can be represented at a five-kilometer resolution.
Other surface variables are produced at ten kilometers, while atmospheric variables such as wind use a 25-kilometer scale.
The previous WeatherNext 2 system used a 25-kilometer grid with six-hour updates, according to Google.
The research paper describes the model as a Functional Generative Network mesh transformer.
Its outputs include dense gridded fields, tropical cyclone tracks, and station-level predictions.
The station output can estimate two-meter temperature and dewpoint at a location and time.
1,000+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
1,000+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
The model uses local geographic features when producing those station-level predictions.
This design tries to connect broad atmospheric patterns with conditions measured at specific places.
It also moves some forecast processing away from a strict sequence of data assimilation, prediction, and post-processing stages.
The paper says WeatherNext 3 predicts satellite-derived precipitation estimates, cyclone information, and station observations.
These inputs provide a more direct connection to observed conditions than training only on model-generated analysis fields.
The authors report that the model achieves leading probabilistic medium-range forecast performance.
That conclusion comes from the authors’ evaluation and requires careful interpretation.
Google also cites independent live evaluations by Brightband in its announcement.
The public material does not make Brightband’s evaluation a replacement for weather-agency verification.
Forecast skill varies by variable, location, lead time, observation quality, and the event being predicted.
A higher-resolution output can show more local detail without eliminating uncertainty.
Hourly updates can also create a false sense of precision if the underlying observations are sparse or noisy.
The model is being integrated into Google Search, Gemini, Maps, Google Maps Platform, and Cloud, according to the announcement.
That distribution gives the research a practical deployment path beyond an academic demonstration.
It also raises a separate question about how users see uncertainty when a forecast is embedded inside a general-purpose product.
The model’s research paper makes several claims about performance, but it does not establish perfect forecasts for local events.
It also does not show that AI forecasting should replace national meteorological agencies or physics-based models.
The paper instead presents WeatherNext 3 as a system that combines learned atmospheric patterns with recent observations.
That combination addresses one weakness of earlier AI systems that inherited delays and biases from analysis data.
The operational importance is therefore not only model size.
It is the attempt to reduce observation latency while preserving global coverage and probabilistic output.
WeatherNext 3 shows how AI forecasting is moving toward systems that ingest measurements continuously.
Its useful test will be sustained comparison across weather conditions, regions, variables, and warning decisions.
Quantitative Thinkers Are Building Ideas Worldwide on WorldQuant BRAIN
WorldQuant BRAIN is a global simulation and research platform for students, aspiring researchers and quantitative thinkers from around the world.
On BRAIN, that spirit of exploration continues through challenges, collaboration and opportunities to develop new quantitative research ideas.
