Ask Cat › AI Tool Summary › Gemini
Google's WeatherNext 3 Is Live: Hourly Forecasts at 5km, and Four Ways Developers Can Pull the Data
Article last updated:2026-09-06
On 2026-09-03, Google DeepMind and Google Research launched WeatherNext 3. It reads like science news, but for anyone building products the important part is the last paragraph: four developer access channels opened at the same time.
1. What the official page states
Reading Google DeepMind’s official page directly:
- Forecasts every hour of the day
- Weather-station-targeted temperature and humidity at 5km resolution
- Other surface variables like wind at 10km
- Already delivered into Google products: Search, Maps and Gemini
Press coverage adds that atmospheric variables such as wind speed run at 25km, and that precipitation forecasts a day or more ahead are up to 50% more accurate, with the largest gains in regions where forecasts were historically less reliable. Neither figure appears on the official page, so we attribute both to media reporting.
2. The four developer channels
| Channel | Best suited to |
|---|---|
| Google Maps Platform Weather API | Querying weather inside an app or site |
| BigQuery | SQL analysis over large forecast and historical datasets |
| Google Earth Engine | Geospatial analysis, layering with other datasets |
| Google Cloud Storage | Bulk download (media reporting says Zarr format) |
The point of that combination: you do not have to run the model yourself. Query live through the API, analyse through BigQuery or Earth Engine, or build your own pipeline from bulk files. For products in agriculture, logistics, events, energy or insurance, this is a data source you can wire in directly.
3. Three things to watch
1. The official page states no pricing. All four channels are billable Google Cloud services, and real cost depends on query volume and method. The WeatherNext page itself says nothing about price or free quotas — check each service’s own pricing page.
2. Resolution is not one number. 5km applies to station-targeted variables like temperature and humidity; other surface variables such as wind are 10km. Check which variable your application actually depends on.
3. The accuracy gain is relative. “Up to 50% more accurate” is an improvement against a baseline, not an absolute hit rate. For decisions that matter, backtest against your own region.
4. What it means if you do not code
It reaches you regardless — the weather results in Search, Maps and Gemini are now produced by this model. You do nothing, and the engine behind the numbers has already changed.
Practically, the improvement is most visible when you look a day or more ahead. Same-day short-range forecasts were already reasonably reliable, so the difference there is smaller.
5. Still unverified
- Official launch date: not printed on the official page; 2026-09-03 comes from media reporting. Unverified.
- Actual cost and whether any free quota exists on each channel: not on the official page. Unverified.
- The 25km atmospheric resolution and the 50% precipitation improvement: media-sourced. Unverified.
Verified 2026-09-06. Sources: Google DeepMind WeatherNext page, TechCrunch. Specifications can change; the official page is authoritative. For how Gemini’s plans differ, see our Gemini tool page.
What Amo and Pimi think
Those already in the Google ecosystem: try the free version first, and if it's not enough, AI Plus for US$4.99 is the cheapest paid plan on the site. However, don't subscribe through one.google.com for US$9.99, or you'll be overpaying.
Let's take a look at these
- Gemini Comprehensive Introduction: Pricing, Features, and Actual Limitations
- Gemini Is the free quota enough?
- Gemini Alternatives
- Comprehensive Free Quota List for All Tools

