Google DeepMind and Google Research released WeatherNext 3 on 3 September, a global model generating hourly forecasts that went straight into Search, the Gemini app, Maps, the Maps Platform Weather API and Earth Engine. The two numbers carrying the coverage — five times sharper, 60% better at rain — are both narrower on the page than in the headlines.
"Five times sharper" is a resolution claim
The post describes forecasts "roughly five times sharper than our previous model, WeatherNext 2, which produced forecasts on a 25-kilometer grid in 6-hour increments." Sharpness here is grid spacing, not skill — resolution says nothing on its own about whether a forecast is right. The cadence change is real and arguably the bigger operational shift: a new forecast every hour, against a six-hour data lag in conventional numerical weather prediction.
The 5 km figure covers two variables
The resolution is not uniform, and the post is explicit: temperature and moisture at 5 km, "other surface variables at 10 kilometers, and atmospheric variables, like wind speed, at 25 kilometers." Wind aloft therefore sits on the same grid as WeatherNext 2. Headlines reading "hourly 5-kilometer forecasts" describe the best case as though it were the whole model.
Three baselines, and the coverage picks one
On precipitation the post reports "a Continuous Ranked Probability Score (CRPS) improvement of up to 60% against IMERG, 30% for MRMS, and 10% against rain gauge measurements for early lead times." IMERG is a satellite precipitation estimate; rain gauges are the closest thing available to ground truth, and that is the smallest of the three by a factor of six. The separate "up to 50% more accurate" figure is a consumer-product claim about Search and Maps when planning a day ahead, not a model benchmark.
What the received framing gets wrong
Reported as "Google's most accurate weather model", the release does not make that claim on its own authority. The post attributes the ranking to a third party — "the most advanced and accurate global weather model to date, according to independent live evaluations by Brightband" — and carries no comparison against ECMWF's operational forecast anywhere, which is the baseline the claim would normally require. New variables aimed at renewables operators, including 100 m wind speed, high-resolution cloud cover and solar radiation, are the substantive addition.