The ECMWF AIFS weather model

At a glance

What ECMWF AIFS is

A physics-based model such as the IFS calculates how the atmosphere should change from one step to the next. The AIFS learns that from history instead: it is trained on decades of past atmospheric states, and ECMWF describes it as the first fully operational open weather prediction model that uses machine learning.

The result is a model that is quick and far cheaper to run. ECMWF reported at launch that it used about a thousand times less energy for a forecast, and that it outperformed physics-based models on many measures, including tropical cyclone tracks, with gains of up to 20 percent. For the ensemble version, ECMWF reported better skill than the physics-based ensemble in the medium range, with surface temperature gains of up to 20 percent.

ECMWF upgraded both versions on May 12, 2026, adding wave and snow cover forecasts and improving the stratosphere. ECMWF also notes poorer 2 meter temperature forecasts in the Arctic, most noticeable in winter, so this is a model with strengths and gaps like any other.

Where it does well

• Large-scale pattern and cyclone track. ECMWF reports gains over physics-based models on many measures, including tropical cyclone tracks.
• Speed. It produces a forecast in a fraction of the energy and time.
• Four runs a day to 15 days, so a new look arrives often.
• A different method, so agreement with a physics-based model comes from independent approaches.

Where it falls short

• Resolution. It is coarser than the 9 km IFS, so it does not draw fine detail.
• It lacks the explicit physics of convection and precipitation, so treat precipitation amounts and sharp features with caution.
• ECMWF lists specific weak spots in version 2, such as 2 meter temperature in the Arctic winter.
• It is new, and forecasters have less experience with it, so use it as a second opinion.

What it is typically used for

• A second opinion on storm tracks and the large-scale pattern, especially for tropical cyclones and coastal lows.
• Checking whether a shift in a physics-based model has support.
• Comparing an AI model and a physics-based model at the same valid time.

How to see ECMWF AIFS in 4070

ECMWF AIFS (AI) is in Model Explorer under Global Models with runs at 00Z, 06Z, 12Z and 18Z. It is part of 4070 Pro, which starts with a seven-day free trial.

• Fields include 2 meter temperature with MSLP, 10 meter wind with MSLP, precipitation with MSLP, 850 mb temperature with MSLP and precipitation, a classic thickness and MSLP view, 850 mb temperature with heights and winds, 500 mb heights and wind, 250 mb winds and heights, and 2 meter dew point.
• The map region is North America.
• Plumes include four recent AIFS runs at a U.S. point and the AIFS ensemble as a separate group of members. The AIFS ensemble appears in Plumes and Storm Tracks and not in Model Explorer.
• Ensemble storm tracks include AIFS ENS for tropical systems. The National Hurricane Center remains the official source.

How it fits with the other models

The IFS is ECMWF’s physics-based model and the AIFS is its machine-learning counterpart, so comparing them at the same valid time is a clean test of method. NOAA AI-GFS is the American machine-learning model, and the GFS is the physics-based American model. When all four agree, that is hard to ignore. When they split along the line of method, that is worth noticing too.

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.