If your business loses money when the weather turns — a crew stood down, a delivery re-routed, a pour of concrete, a wedding marquee, a harvest, a crane lift — something quietly changed in the tool you already use. Google's weather forecast now rewrites itself every hour, and it is free in Search and Maps.
Here is what actually happened, what it is good for, where it still gets things wrong, and how a small business can use it without buying anything.
What Google actually changed
On September 3, 2026, Google DeepMind announced WeatherNext 3, the newest version of its AI weather model. Two things are different, and only one of them is the obvious one.
The obvious change is accuracy. The more interesting change is freshness.
To understand why, you need one piece of background, and it is not technical. Most AI weather models do not look at the weather. They look at a tidy summary of the weather called a reanalysis — a single consistent global snapshot built by blending every available measurement together, including estimates for the huge parts of the planet where nobody is measuring anything. That snapshot is normally produced only every six hours.
So the model was always working from a picture of the sky that could be up to six hours old. As Ars Technica put it in its September 8, 2026 write-up of the technical paper, the raw data sources "may contain some information that gets lost in the reanalysis blender."
WeatherNext 3 now also reads raw weather satellite imagery directly, as it arrives. Because it no longer has to wait for the six-hourly summary, it can produce a brand new forecast every hour. Ars Technica describes the same change in its own words: the model is "adding in weather satellite data and upping the forecast frequency to hourly as a result."
Google's previous model, WeatherNext 2, produced forecasts on a roughly 25-kilometer grid in six-hour steps. Google says the new one resolves temperature and moisture down to about 5 kilometers, other surface conditions to about 10 kilometers, and upper-air conditions like wind speed to about 25 kilometers.
Why "every hour" matters more than "more accurate"
Business owners hear "more accurate forecast" constantly and have learned to ignore it. Freshness is a different promise, and it is the one that shows up in your day.
Weather that costs you money usually arrives fast. A line of storms forms in an afternoon. A front moves in ahead of schedule. Fog settles. Under the old cycle, the model's view of that developing situation could be hours stale by the time you looked at your phone. Under the new one, the forecast you are reading was built from satellite images taken within the last hour or so.
On top of that, Google claims a real accuracy gain. In the technical paper, Google reports roughly a 5 percent improvement in upper-atmosphere accuracy over its previous model — which, Google says, equates to about six more hours of usable forecast lead time. Ars Technica read the paper and reported that figure, noting it is Google's own characterization of its own results.
Six hours is not an abstraction if you run anything outdoors. Six hours is the difference between canceling a job the night before and canceling it when the crew is already on site with the truck loaded.
Two other numbers Google puts forward, both of which are the company's own measurements rather than an outside audit:
- Precipitation. Google says people planning a day or more ahead will see up to 50 percent more accurate rain and snow forecasts, with the biggest gains in regions where forecasts have historically been least reliable.
- Local temperature. The model now factors in whether a specific point is land or ocean and how high above sea level it sits. Google reports that this improved accuracy for a specific location by up to 30 percent — which matters if you sit in a valley, on a coast, or in the mountains, where conditions a few miles away are genuinely different from yours.
Where to get it, and what it costs
This is the unusual part: for most business owners, the answer is nothing and nowhere.
Google says WeatherNext 3 started powering the weather you already see in Google Search, Google Maps, and the Gemini app from the day of the announcement. There is no product to buy, no signup, no app to install, and no setting to change. If you type "weather" into Google or check the weather layer in Maps, you are already looking at the new model's output.
If you want the underlying data rather than the consumer view — feeding forecasts into your own scheduling system, dispatch tool, or spreadsheets — Google routes that through its commercial platforms: the Google Maps Platform Weather API, plus BigQuery, Earth Engine, and Google Cloud Storage for bulk forecast data. Google's developer documentation notes that the bulk data channels currently require requesting allowlist access, and it does not publish a price on that page. Treat that path as a real project with a real bill attached, and get a quote before you plan around it. For most small businesses, the free version in Search and Maps is the whole story.
Where it still gets things wrong
A forecast is a probability, not a promise, and this model has documented quirks. Ars Technica flagged several that are worth knowing before you lean on it:
- It can be worse in the very short term. For a number of variables, WeatherNext 3 performed worse than competing models on the first six-hours-ahead forecast, before pulling ahead for the rest of the 15-day window. Ars notes the paper does not even offer a guess as to why. In other words: for "what is happening this afternoon," it is not automatically the best answer available.
- Visible artifacts. Some precipitation maps show distinctly hexagonal blobs — the shape of the model's own internal grid showing through the output rather than real weather.
- It is a black box. Unlike a traditional forecast, which simulates the physics of the atmosphere, an AI model learns patterns from past data and produces a prediction. When it is wrong, there is usually no explanation you can inspect.
Google is explicit about the limit, too. Its own post carries a disclaimer: for official forecasts, severe weather warnings, and public safety advisories, use your local meteorological agency or national weather service. That is the right rule. Use this to plan your week; use the National Weather Service to decide whether it is safe to send people out.
How to use it this week
No budget, no software, no consultant. This is a habit change, not a purchase.
- Re-check before you commit, not just the night before. The old habit of checking once at 6 p.m. for tomorrow made sense when the forecast only updated a few times a day. It no longer does. Check again in the morning, and again before a crew rolls.
- Check the forecast for your actual site, not your nearest city. The resolution improvement is the whole point. Search the job address or drop the pin in Maps.
- Write down your weather trigger. Decide in advance what number moves a job — wind above a certain speed, any accumulation, a temperature floor for a pour or a paint job. A specific threshold turns a forecast into a decision instead of an argument.
- Push the cancellation decision later on purpose. If forecasts are fresher and hold up further out, the value comes from waiting for better information rather than committing early out of habit. That only works if your customers know your call time in advance, so put it in writing.
- Keep the official source for safety calls. Warnings, watches, and advisories come from your national weather service. Nothing here replaces that.
The bigger pattern
This is worth noticing even if you never think about the weather. Nobody sold you WeatherNext 3. Nobody asked you to adopt it, train on it, or budget for it. An AI model was swapped in underneath a tool you have used for fifteen years, and the tool got better on a Tuesday.
That is increasingly how AI reaches small businesses: not as a product you evaluate, but as a silent upgrade inside something you already use. The upside is that you get the benefit for free. The catch is that the limits arrive silently too — the quirks above are real, and they are not mentioned anywhere on the screen where you read the forecast. Knowing what is underneath the number is the difference between using a tool and trusting it blindly.
We break down developments like this one for people who run businesses, not for engineers. More explainers are in the AIWAS knowledge base, and short takes go out on @AIWASai.