Google DeepMind and Google Research have released WeatherNext 3, a global AI weather model designed to forecast key variables at 5km resolution and refresh predictions hourly. TechCrunch reports that its output will feed Search, Maps and Gemini, bringing the work much closer to the moment somebody decides whether the umbrella is worth the indignity.
Google DeepMind Watch analysis
What happened
WeatherNext 3 was released on 3 September 2026, according to TechCrunch. Google says the larger model can use hourly satellite observations, target individual weather stations and improve its rain evaluations by 60% over WeatherNext 2.
Google also plans to offer the model through its cloud platforms. WittyWires has not independently reviewed a corresponding Google launch document, so the technical and rollout details here are attributed to the company through TechCrunch’s reporting.
Key findings
- Five-kilometre detail
Google says WeatherNext 3 reaches 5km resolution for key variables, narrowing the gap between global forecasting and genuinely local decisions. - Hourly forecasts
Predictions can arrive every hour rather than every six, making sudden changes less likely to spend half a day hiding in the machinery. - Better rain evaluation
Google reports a 60% improvement over WeatherNext 2, addressing one of AI forecasting’s more stubborn weak spots. - Strong benchmark showing
Brightband’s Operational WeatherBench reportedly placed the model ahead of the AI and conventional systems included in its tests. - Rawer observations, with caveats
WeatherNext 3 ingests hourly satellite data directly, but still depends partly on national weather datasets for its forecasts.
Why it matters
Finer, more frequent forecasting could improve ordinary decisions in Google’s mass-market products, but the larger prize sits beyond picnic logistics. Better wind, rain and cloud predictions could help airports, energy operators and regions without lavish forecasting infrastructure plan with greater confidence.
The important qualification is that benchmark leadership is not the same as universal superiority. Extreme events, different regions and messy real-world deployment have a habit of turning victory laps into field tests.
Our read
WeatherNext 3 looks like a meaningful step because Google intends to put it where people already check the weather, not leave it admiring itself in a research paper. The useful test is now brutally simple: do those finer forecasts produce noticeably better decisions when the sky stops cooperating?
What to watch
- Which countries and Google products receive WeatherNext 3 data first.
- How it performs on heavy rain, rapidly developing storms and other costly edge cases.
- Whether independent forecasters reproduce its reported benchmark advantage.
- How far Google can reduce the model’s remaining dependence on government-processed datasets.
Discussion spark: Would more local, hourly AI forecasts change how much you trust Google’s weather information, or do extreme-weather misses matter more than everyday accuracy?
Sources and evidence
- Google's latest AI weather model gives you no excuse to forget your umbrella | TechCrunch (3 September 2026, 15:00 UTC)
not affiliated with, endorsed by, or operated by Google or Google DeepMind
