The NOAA AI-GFS weather model

At a glance

What NOAA AI-GFS is

Most weather models solve the equations of fluid motion on a grid, step by step. A machine-learning model works differently: it learns from decades of past atmospheric states how the atmosphere tends to evolve, and then applies that learning to produce the next step. NOAA’s AI-GFS is based on GraphCast, a published machine-learning weather model, adapted by NCEP, NOAA research labs and EPIC.

The practical result is speed and cost. NOAA says a 16-day AIGFS forecast uses about 0.3 percent of the computing of a GFS run. That makes it a cheap second opinion on the large-scale pattern, which is where it earns attention.

NOAA released it alongside an ensemble, AIGEFS, and a hybrid, HGEFS. In 4070, the deterministic AI-GFS is in Model Explorer and Plumes.

Where it does well

• Large-scale pattern. NOAA reports better skill than the GFS for many large-scale features.
• Tropical cyclone tracks. NOAA reports better tracks than the GFS for tropical cyclones.
• Speed and cost. Because it is cheap to run, it can add a meaningful second opinion without waiting for a big computer.
• A genuinely different method. When a physics-based model and a machine-learning model agree, the agreement comes from two independent approaches.

Where it falls short

• Intensity. At launch NOAA said intensity forecasts needed improvement. Version 1.1, operational July 27, 2026, targets intensity and precipitation, and it has had little time to prove itself.
• Coarser time steps. Output comes every 6 hours, so it cannot show hourly detail.
• A smaller set of fields. It does not produce everything the GFS does, so some maps simply are not available.
• It is new. Forecasters have had less time to learn its habits, so read it as a second opinion, not as the deciding model.
• A model that learns from past weather can be less reliable on rare extremes. Check an outlier against a physics-based model.

What it is typically used for

• A second opinion on the large-scale pattern and storm track beside the GFS and the ECMWF IFS.
• Looking at 500 mb heights and wind, temperature and precipitation out to 16 days.
• Checking whether a track shift in one model has company.

How to see NOAA AI-GFS in 4070

The model appears as NOAA AI-GFS (AI) in Model Explorer under Global Models, and as AIGFS in Plumes. It is part of 4070 Pro, which starts with a seven-day free trial.

• Fields include 2 meter temperature with 10 meter winds, precipitation with MSLP, 6-hour, 24-hour and total precipitation, 1000 to 500 mb thickness, 500 mb temperature and heights, 500 mb heights and wind, 850 mb temperature and heights, and 250 mb winds and heights.
• The map region is the United States.
• Plumes include the four most recent AIGFS runs at a U.S. point, beside the physics-based models.

For more on 500 mb, see how to read a 500 mb map.

How it fits with the other models

The GFS is NOAA’s physics-based global model, and the AI-GFS is its machine-learning counterpart. The ECMWF AIFS is the European machine-learning model. Reading all four of the GFS, AI-GFS, ECMWF IFS and AIFS side by side at the same valid time is a good habit, and the guide on model accuracy explains why. For confidence, add an ensemble such as GEFS.

Specifications change when agencies upgrade their models. The Sources below link to the producing agency. Official National Weather Service forecasts, watches and warnings remain the authority for what is expected where you live.