Sunny reads ten forecast models and seven weather services. That number isn't the point — plenty of apps advertise big source counts and quietly average them into mush. The point is the test each one had to pass: it has to add independent signal, not repeat something already in the room. A service that just re-blends models we already read would inflate agreement without adding information, so we leave those out — and when sources disagree, you see the disagreement.
Numerical weather models are the raw computation — government supercomputers (and now one AI) simulating the atmosphere from scratch several times a day. Every weather app on Earth is built on these. Sunny reads them directly, unaveraged, and the chart shows each one as its own line.
The Global Forecast System — the workhorse American global model, run four times a day and published fully open, which is why half the weather apps on Earth are built on it.
Why it's hereThe reference point. When another model disagrees with GFS, that gap is usually the day's real story.
The Integrated Forecasting System from the European Centre for Medium-Range Weather Forecasts — the model that tops medium-range verification scores more often than any other.
Why it's hereWidely regarded as the world's most accurate global model. A consensus without it would be negligent.
Germany's ICON (icosahedral nonhydrostatic) model — a different grid geometry and different physics choices from the American and European mainlines.
Why it's hereGenuinely independent engineering. Its disagreements with GFS and ECMWF are informative precisely because it isn't derived from either.
The High-Resolution Rapid Refresh — a 3 km storm-scale model covering the continental US, re-run every hour with fresh radar folded in.
Why it's hereFor the next few hours over the US, nothing global competes. It self-gates: outside its domain it simply doesn't vote.
The Met Office's Unified Model — one codebase spanning global forecasting to UK regional detail, developed continuously since the 1990s.
Why it's hereA top-tier center with its own lineage. Adds an independent voice that frequently splits the difference between the American and European calls.
The Global Environmental Multiscale model — Canada's own global system, with particular strength in the patterns that matter at northern latitudes.
Why it's hereIndependent physics and an agency with a home-field advantage on cold-season weather, which is where forecasts disagree most.
France's ARPEGE global model and its high-resolution AROME companion over Europe, run by one of the oldest national weather services in the world.
Why it's hereAnother fully independent center — and over western Europe, AROME's fine grid picks up detail the global models smooth away.
The Japan Meteorological Agency's global model, from the agency that pioneered much of modern typhoon forecasting.
Why it's hereAn independent Pacific-facing perspective. On systems crossing the Pacific toward North America, JMA often leads the pack.
The China Meteorological Administration's global model — a fully independent forecast center: its own physics, its own data assimilation, not a re-blend of anything else in this list.
Why it's hereHonesty about the range of official world models means not curating away the ones that stray. CMA verifies below the majors, and on its stray days the spread widens — the accuracy scorecard exists to show exactly that, in public.
The Artificial Intelligence Forecasting System — ECMWF's machine-learned model, trained on decades of reanalysis rather than solving physics equations. The first non-physics voter in the row.
Why it's hereIt verifies competitively with the flagship physics models, and it fails differently — its errors are uncorrelated with everyone else's, which is precisely what a consensus wants from a new voice.
Services sit on top of the models — agencies and companies that blend, correct, and post-process them into the forecasts people actually consume. Each votes in Sunny's consensus, and the median across them is the bright line in the chart. Where a service is proxied, your coordinates travel via our servers and your IP address never reaches the provider.
The aggregation backbone: an open API serving current conditions, hourly and daily forecasts, and — crucially — the per-model comparison that powers the chart and the model vote above.
Why it's hereIt's the only service that hands over the raw models unaveraged, which is the whole product. Sunny pays for its commercial tier.
Fetched via our servers; coordinates are rounded before they leave.
The Norwegian Meteorological Institute's public forecast API — the same data behind Yr, one of Europe's most used weather sites.
Why it's hereA national agency publishing its own post-processed forecast globally, for free, as public infrastructure. Independent of every commercial blend here.
Fetched directly from your device.
The US National Weather Service — official government forecasts, and the source of every severe-weather alert Sunny shows in the United States.
Why it's hereIn the US it is the forecast of record, written by human forecasters. Alerts from it are never paywalled and never blended with anything.
Fetched directly from your device. US-only; elsewhere it simply doesn't vote.
Apple's weather service, descended from Dark Sky, blending global models with Apple's own processing.
Why it's hereOne of the most-consumed forecasts on Earth — when your phone's default app disagrees with the model row above, this voter is why you'll see it.
Proxied through our servers — Apple never sees your IP address.
An open reimplementation of the Dark Sky API, built transparently on public model data by an independent developer.
Why it's hereA blend whose recipe is published — the only voter here whose post-processing you can actually read. Transparency earns a seat.
Proxied through our servers — the provider never sees your IP address.
One of the longest-running commercial weather APIs, serving a proprietary blend at three-hour steps.
Why it's hereA widely-consumed independent blend with two decades of operational history. Its coarser cadence is visible in the chart — we show it rather than smoothing it.
Proxied through our servers — the provider never sees your IP address.
Google's weather service, which layers Google's machine-learning forecasting on top of the physical models.
Why it's hereLike Apple, a forecast consumed by billions — and its ML layer means it genuinely diverges from the raw models it's built on.
Proxied through our servers — Google never sees your IP address.
We evaluate new sources regularly, and most fail the same test: they resell a blend of GFS and ECMWF — models Sunny already reads directly — so adding them would count the same opinion twice and make the consensus look more certain than it is. The other disqualifiers: terms that forbid commercial use or caching, and official warnings that arrive altered. Every source above is also graded nightly against what actually happened at the nearest weather station; months of that history become the accuracy scorecard in the app.