WeatherNext 3 ships hourly 5km global forecasts to devs via BigQuery
Refreshes hourly out to 15 days with a 64-member ensemble and up to 50% better rain calls. Plus: Meta's Muse Spark 1.3 reasoning model lands on OpenRouter.
Copy markdown
Hourly refresh, 15-day reach
WeatherNext 3 re-initializes every hour and forecasts 15 days out in 1-hour steps at ~5km surface resolution, with a 64-member ensemble — roughly 5x finer than WeatherNext 2's 25km, 6-hour cadence.
How you actually call it
No plain REST endpoint yet: you pull forecasts from BigQuery and Earth Engine, bulk-download ensembles as Zarr from Cloud Storage, or hit the Google Maps Platform Weather API. Output spans 40+ variables including solar irradiance and wind.
The accuracy jump, in numbers
Google reports up to 50% more accurate precipitation beyond day one, 30-40% CRPS gains on short-range 2m temperature, and about 10% over ECMWF's AIFS ENS v2 on upper-air variables in forecast week one.
What you could ship this weekend
A queryable, hourly-updating global forecast is a new building block: wire it into a solar/wind generation scheduler, a rain-aware logistics or events agent, or a farm-ops dashboard — the built-in irradiance and wind fields are made for exactly that.
Elsewhere: Meta's Muse Spark 1.3 hits OpenRouter
Meta's new reasoning model is live on OpenRouter at $1.25/$4.25 per million tokens with a 1M context and text, image and video input — scoring 62 on Artificial Analysis's Intelligence Index, well above its price tier.