The Global Forecast System (GFS) is the National Weather Service's main global weather model, run by the National Centers for Environmental Prediction (NCEP). It forecasts the entire planet four times a day out to 16 days, which makes it a core tool for the large-scale pattern beyond the next couple of days.
How it is built
As of October 2026, the operational version is GFS version 16, which entered operations on 03/22/21. It runs on the Finite-Volume Cubed-Sphere (FV3) dynamical core, the part of the model that solves the equations of motion, at about 13 km (8 miles) resolution with 127 vertical layers.
Runs start at 00, 06, 12 and 18 UTC. Output is hourly for the first 120 hours, then every 3 hours from day 5 to day 16, a 384-hour run.
Where each run starts
Each run begins from the Global Data Assimilation System, which blends observations from around the world into a starting state using a hybrid ensemble-variational method. The atmosphere is coupled to NCEP's global wave model.
How forecasters use it
Forecasters use the GFS to track troughs, ridges, jet streaks and surface lows days ahead, then compare it with other global models and ensembles. It also feeds other guidance.
The National Blend of Models uses the GFS among its inputs, and the new Rapid Refresh Forecast System takes its lateral boundary conditions, the weather flowing in at its edges, from the GFS.
Limits and what is next
At 13 km the GFS is too coarse to simulate thunderstorms directly, so convective details need a convection-allowing model. Small errors in the starting data also grow with time, so a single run's day 10 to 16 details carry little weight on their own.
In April 2026 NOAA proposed GFS version 17, a coupled ocean, sea ice, wave and atmosphere system at about 9 km, for implementation in October 2026. If it is implemented, the resolution and setup described here will change.