Model profile
The GFS weather model, explained
Last updated: 14 November 2026
The Global Forecast System, or GFS, is the United States’ global weather model. It is free, it updates four times a day and it forecasts 16 days ahead, which is why it sits behind a large share of the world’s weather apps and websites. This guide covers how it works, what it is good at and when to be sceptical of it.
GFS at a glance
| GFS | |
|---|---|
| Full name | Global Forecast System |
| Run by | NOAA / National Centers for Environmental Prediction (USA) |
| Resolution | ~13 km, 127 vertical levels |
| Runs | 00 / 06 / 12 / 18 UTC |
| Forecast range | 384 h (16 days) |
| Ensemble | GEFS, 31 members |
| Data | Public domain, free |
What the GFS is
The GFS is produced by the National Centers for Environmental Prediction (NCEP), part of the US National Weather Service and NOAA. It is a global, physics-based model: it covers the whole planet, from the surface to the upper atmosphere, and steps the equations that govern the atmosphere forward in time from an analysis of current conditions.
Because it is paid for by US taxpayers, its output is in the public domain. Anyone can download every run at no cost and with no licence restrictions. That one policy decision has made the GFS the most widely used weather model in the world.
How it works
Since 2019 the GFS has been built on a dynamical core called FV3, which represents the globe as a cube projected onto a sphere. This avoids the shrinking grid cells that a latitude–longitude grid has near the poles. A 2021 upgrade raised the number of vertical levels to 127 and extended the model higher into the atmosphere.
Each run starts from the Global Data Assimilation System, which merges satellite, aircraft, balloon, radar and surface observations with the previous forecast. The model then produces hourly output for the first five days and three-hourly output after that, all the way to 384 hours.
When the GFS updates
A new run starts every six hours, at 00, 06, 12 and 18 UTC, and every one of them goes the full 16 days. Output starts to appear about three and a half hours after the start time and the run is complete roughly an hour and a half later.
Four full-length runs a day is the GFS’s biggest practical advantage. ECMWF’s main runs are twice a day and arrive later, so for much of the day the newest global guidance available is a GFS run.
What the GFS is good at
The GFS is a strong all-round model for the first five to seven days. It handles the large-scale pattern — the position of highs, lows, fronts and the jet stream — well, and its hourly output in the first five days makes it useful for timing.
It is also the only widely available global model that runs to 16 days four times a day, so it is the main tool for spotting pattern changes in week two: a cold outbreak, a heat dome building, a shift to a wetter regime.
Where to be sceptical
Averaged over years, the GFS trails ECMWF by about a day of forecast skill in the medium range, and usually sits a little behind the UK Met Office model as well. The gap has shrunk since the FV3 upgrades, but it has not closed.
The second caution is the long range. Beyond about day 10, individual GFS runs can swing wildly from one run to the next — a major snowstorm in one run, nothing in the next. Forecasters sometimes call this the “fantasy range”. A single run at day 12 is one possible outcome out of many, not a forecast to plan around.
For that range, the right tool is an ensemble.
GFS vs GEFS: what is the difference?
The GFS is one forecast: a single run from the best estimate of current conditions. The GEFS, the Global Ensemble Forecast System, runs the same model 31 times at lower resolution, each from slightly different starting conditions.
Use the GFS for detail in the first few days. Use the GEFS to judge confidence, especially in week two: if most members agree, the signal is real; if they scatter, the single GFS run you are looking at is close to a coin toss. The “spaghetti plots” shared during hurricane season are ensemble members and other models drawn on one map for exactly this reason.
NOAA’s AI version
Since December 2025 NOAA has also run an AI-based global model alongside the GFS, known as the AIGFS. It is built on Google DeepMind’s GraphCast and fine-tuned on NOAA’s own analyses, and it produces a 16-day forecast with a small fraction of the computing power the GFS needs. It is run next to the physics-based GFS, not instead of it.
How to use the GFS well
Treat the GFS as a fresh, capable opinion and check it against the others. Where the GFS, ECMWF and ICON agree, confidence is high. Where the GFS stands alone — especially with a dramatic storm a week or more out — wait for the next run and for the other models.
On the MeteoModels map you can overlay the GFS with the other global models and turn on the certainty mask to see, for any place and time, whether the GFS has company.
Frequently asked questions
What does GFS stand for?
GFS stands for Global Forecast System. It is the global weather model run by NOAA’s National Centers for Environmental Prediction in the United States.
How often does the GFS update?
Four times a day, at 00, 06, 12 and 18 UTC. Each run forecasts the full 384 hours (16 days) ahead.
How far ahead does the GFS forecast?
384 hours, or 16 days. Accuracy drops steadily with lead time, and individual runs beyond about day 10 should be read as possibilities, not forecasts.
How accurate is the GFS?
It is a reliable model for the first five to seven days and one of the leading global models. On average it trails ECMWF by about a day of forecast skill in the medium range.
What is the difference between GFS and GEFS?
The GFS is a single high-resolution forecast. The GEFS is an ensemble of 31 lower-resolution forecasts from slightly different starting conditions, used to measure how confident the forecast is.
Is GFS data free?
Yes. GFS output is in the public domain and free for anyone to download and use, which is why so many weather apps are built on it.